The base model M4 Mac mini is one of the best values in Mac computers, but upgrading the storage to the maximum 2TB significantly reduces that value proposition. That’s because Apple charges an insane $800 for the upgrade, turning the budget $600 computer into a $1400 one. In other words, you can purchase two base model Mac minis for less than the price of a single Mac mini with a 2TB upgrade. But fret not, because Expand Mac mini offers a 2TB upgrade for the M4 Mac mini for just $320 – less than half of what Apple charges. In this hands-on step-by-step video walkthrough, I show you how to upgrade the M4 Mac mini SSD with 2TB of storage. Watch my full video, and be sure to subscribe to 9to5Mac on YouTube for more Mac guides and tutorials. Disclaimer: This only works with the Mac mini with M4 chip. It does not work with the higher-end M4 Pro Mac mini. Things you’ll need: Base M4 Mac mini Expand Mac mini 2TB SSD A second Mac running macOS Sonoma 14 or later for restoring macOS External SSD for Time Machine backup and restore A USB-C to USB-C cable that supports data and charging, not a Thunderbolt cable T3, T5, and T8 Torx drivers Jimmy tool Opening Pick Tweezer (Optional) Spudger (Optional) Expand Mac mini includes a multi-bit screw driver, the necessary bits, and pryer tool, but I much prefer my iFixit Pro Tech Kit, which includes higher quality parts. It’s not a requirement, but if you plan on doing any sort of DIY stuff, this kit is nice to have. Video: Upgrade the M4 Mac mini from 256GB to 2TB Subscribe to 9to5Mac on YouTube for more videos Backup your Mac Before you do anything, be sure to back up your Mac’s current SSD using Time Machine. This will let you easily restore your machine using Migration Assistant once the new SSD is installed and DFU-recovered. Step 1: Connect an external SSD with free space. Time Machine backups require their own volume, so you can create a separate volume via Disk Utility, or just use the entire SSD. I recommend any manner of Samsung SSDs; I use the Samsung T9. Step 2: Open Settings → General → Time Machine and click Add Backup Disk. Step 3: Enter your administrator password. Step 4: Select your external SSD and choose Set Up Disk… Step 5: Make sure Encrypt Backup is selected, and then enter and verify a new encryption password. Apple also requires you to fill out the “hint” field. Click Done to save the changes and the Time Machine backup will start preparing. Step 6: The Time Machine backup will automatically commence after 60 seconds, or you can force it to start by clicking Time Machine in the menu bar and selecting Back Up Now. Step 7: Verify that the backup is completed by visiting Settings → General → Time Machine. Step 8: Eject the Time Machine SSD from your Mac to save for later use. Erase All Content and Settings This is an optional step, but since you’ll no longer be using the SSD that you plan to replace, I recommend deleting the drive before removing. Ensure that you have your data backed up first before proceeding. Step 1: Go to Settings → General → Transfer or Reset and click Erase All Content and Settings. Step 2: Enter your Mac’s administrator password and click Unlock. Step 3: Click Continue on the Time Machine prompt, since you should have already backed up your Mac. Step 4: Click Continue on the Erase All Contents & Settings page. Step 5: If you’re signed in to iCloud, enter your Apple Account password to sign out and disable Activation Lock. Click Continue. Step 6: Click Erase All Content & Settings. The SSD will be erased, and the Mac will reboot. Install the new SSD Step 1: Shut down your M4 Mac mini and remove all cables. Step 2: Rest the Mac mini face down on a flat, soft surface to avoid marring the aluminum case. Step 3: Carefully insert the Jimmy tool between the gap where the aluminum chassis meets the plastic, black bottom cover, and pry the cover to create a gap between the two. Step 4: Insert a plastic opening pick and remove the Jimmy tool. Step 5: Slide the opening pick around the cover until it encounters each of the four clips that secure the cover to the chassis. Twist the opening pick slightly to pop out each clip. Take special care when sliding around the power button to avoid severing the power button cable. Step 6: Lift the edge of the cover that’s on the opposite corner of the power button, and flip it over to reveal the underside of the back cover. Rest something under the bottom cover so that you can access the inner plate screws without putting strain on the power button cable. Step 7: Remove all eight of the inner plate screws using a T5 Torx driver. All of the screws for the inner plate are the same, so don’t worry about remembering where each screw was. Step 8: Slide the inner plate down and away from the heat sink to loosen it. Be careful, because there is a ribbon cable that connects to the inner plate. Flip the back cover back on top of the inner plate, and then rest both the inner plate and back cover on the side of the Mac mini without putting stress on the ribbon cable. Step 9: Using the T3 Torx driver, remove the four fan screws. There are two short screws on the top, and two longer screws on the bottom. Step 11: Lift the fan and flip it over to expose the ribbon cable connection, and rest it on the edge of the Mac mini chassis without putting strain on the ribbon cable. Step 12: Remove the screw securing the SSD with the T8 Torx driver, and remove the SSD from its socket. Step 13: Insert the 2TB SSD into the socket securely, and secure it using the T8 screw. Step 14: Reinstall the fan, and put the inner plate and back cover back on without securing them down. Before securing the inner plate and back cover, it’s a good idea to test the install to make sure it was successful. DFU Restore Step 1: With the USB-C to USB-C cable, connect to the middle Thunderbolt port on the rear of the M4 Mac mini. This port is designated for DFU restores via macOS. Step 2: Plug the other end of the USB-C cable into a Mac running macOS Sonoma or later. Step 3: Hold the power button on the upgraded Mac mini, and while continuing to hold the power button, plug in the power cable. Step 4: Continue holding the power button until the Mac you’re using to perform the restore recognizes the connected Mac in DFU mode via Finder. You should also notice a flashing orange light on the front of the upgraded Mac mini. If macOS first shows an alert asking you to allow the accessory to connect, release the power button and click Allow. Step 5: Click the Restore Mac button on the Mac DFU Mode Finder window. Step 6: You’ll see a pop up warning you that your Mac will be reset to factory settings. Click Restore and Update. Step 7: macOS will download the necessary software, and then perform the restore. This works similarly to restoring an iPhone via a Mac. Depending on your Internet speed, it could take a while to download the restore file. The restore process is also somewhat lengthy, so be patient. Step 8: Once restored, the upgraded Mac mini’s status light should now be white, and you’ll see a message in macOS on the machine you used to perform the restore stating that the Mac has been restored to factory settings. Click OK. Migrate the Time Machine backup data Step 1: Disconnect your upgraded Mac mini from the other Mac, and connect the upgraded Mac mini to a monitor, keyboard, and mouse. Step 2: Power on the Mac mini, and connect the Time Machine backup drive to one of the USB-C ports. Step 3: Proceed through the initial macOS Setup process until you reach the Transfer Your Data to This Mac page. Select From a Mac, Time Machine, or startup disk, and click Continue. Step 4: Click Not Now on the Accessibility page, and Continue on the Data & Privacy page. Step 5: On the Transfer information to this Mac page, select your external drive, and click Continue. Step 6: If you set the Time Machine backup with encryption, enter the encryption password, and click Unlock. Step 7: Select the Time Machine icon, and click Continue. Step 8: Select your specific Time Machine backup and click Continue. Step 9: On the Transfer Your Information page, ensure that everything you want restored is checked and click Continue. Step 10: Create a new password for the user associated with the Time Machine backup, and click Set Password and then click Continue. macOS will now transfer your information from the Time Machine backup to your upgraded Mac mini’s SSD. This may take a while depending on how large the backup is. Step 11: On the Migration Complete page, click Restart Now to finish the migration. Step 12: Click Done when prompted, and macOS should display the initial Hello screen to finish the remainder of the initial setup. This includes logging back in to iCloud, setting up Disk Encryption, Touch ID, etc. Test out your Mac to make sure everything works, and verify the 2TB SSD upgrade via Finder and/or Disk Utility. Finish Mac mini reassembly If the install went as planned, you may now reassemble the Mac mini’s inner plate and back cover. Use the eight T5 Torx screws to secure the inner plate, and align the clips on the bottom cover to the holes on the inner plate and press down to lock the bottom plate back into place. Congratulations, you just saved hundreds of dollars on a 2TB SSD upgrade! 9to5Mac’s Take The nice thing about this install is that it’s totally non-destructive and can easily be reverted. It also requires no cable removals, or anything else weird. The install requires removing just 13 screws, and can be done in a matter of minutes. I wanted to challenge myself, and I was able to open the Mac mini and perform the install in 5 minutes. It’s the Time Machine backup, DFU restore, and migration that takes up the majority of the time, as the actual hardware install is cake as long as you have the needed tools. 256 GB SSD Speed Test 2TB SSD Speed Test When comparing the two drives, the performance of the upgraded 2TB SSD is predictably faster when it comes to write speeds. But I’m curious to see how the drive holds up over the long term. Are Apple’s drives that much better from a reliability standpoint that it warrants a $500 price difference, or are these drives largely the same, with Apple making insane margins for each upgrade sold? Time will tell, but I’ll be sure to report back with more info once I’ve been able to put the machine through the wringer for a few months. Would you consider this upgrade for your M4 Mac mini? Why or why not? Sound off down below with your thoughts. Add 9to5Mac to your Google News feed.  FTC: We use income earning auto affiliate links. More.You’re reading 9to5Mac — experts who break news about Apple and its surrounding ecosystem, day after day. Be sure to check out our homepage for all the latest news, and follow 9to5Mac on Twitter, Facebook, and LinkedIn to stay in the loop. Don’t know where to start? Check out our exclusive stories, reviews, how-tos, and subscribe to our YouTube channel

We have obviously had quite a bit of Grand Theft Auto 5 over the last decade-plus, from multiple re-releases of the game to, of course, continued and consistent updates what many might call the meat and potatoes of the experience for the vast majority of its lifetime- GTA Online. But go back further in the series’ timeline by just one game, and you arrive at the completely contrasting circumstances of Grand Theft Auto 4. The 2008 open world title was obviously highly acclaimed as well – to the extent that, just like GTA 5, it still stands as one of the best reviewed game of all tine – while it’s also safe to say that it sold a ton of copies as well, even though it could obviously never hope to even come close to GTA 5’s commercial accomplishments. And yet, for all of its critical and commercial successes, once Rockstar Games was done with GTA 4 – which was after two post launch single player expansions, which of course is something that GTA 5 itself never got – Rockstar never looked back at the game again. Interestingly enough, however, recent rumblings have suggested that that could change at some point in the not too distant future. Known Rockstar Games insider Tez2 recently claimed on the GTA forums that a PS5 and Xbox Series X/S port of Grand Theft Auto 4 has been in the works for roughly a year at this point, and that Rockstar likely intends to release it at some point later this year- which of course means that it will release before Grand Theft Auto 6, which was recently officially delayed to May 2026 (and might not necessarily be done with delays, for all we know). Admittedly, this isn’t the first time we have had rumours or speculation about a re-release of GTA 4, and obviously none of them have ever amounted to anything up to this point- but given Rockstar’s recent history, there is precedent to suggest that GTA 4 could once again see the light of day. Rockstar has long been in the habit of outright ignoring its past releases, refusing to offer remasters or remakes or anything resembling a re-release even in the face of overwhelming demand. Until not too long ago, the original Red Dead Redemption was a similar case, but with the acclaimed 2010 Wild West epic having been brought to PC, PS4, and Nintendo Switch over the last couple of years, it now does feel likelier than it ever has that Rockstar might be looking to do something similar with Grand Theft Auto 4 as well. After all, that is another beloved title from that same era of Rockstar games that hasn’t seen the light of day in ages, which in turn has really ratcheted up the demand for the game to come back that much more. But if Rockstar does indeed decide to do a GTA 4 re-release, our hope is it’ll be much more than a simple port. One thing most will likely agree on is that the aforementioned Red Dead Redemption re-release should not serve as a yardstick for Rockstar to measure against, because the bar should really be significantly higher. The Red Dead Redemption port really was in almost all ways just a simply and straightforward port. It offered few upgrades, and the ones that it did were negligible at best. The resolution was upscaled, there were some light improvements to the picture quality, the UI saw some minor adjustments here and there- and that was pretty much it. There were no upgrades to the lighting, to the assets or the textures, or anything of note, really. Hell, there wasn’t even a performance mode- even when played on PS4 Pro or on PS5 via backward compatibility, the game remains locked at 30 FPS, which is just a crying shame, to put it mildly. To top it all off, Rockstar decided to price this re-release – which was deemed by many to be a low effort port of a game over a decade old – at a $50 point, which felt unreasonable, to say the least. Thankfully, the package did include the Undead Nightmare expansion for that price, so at least there was no additional fleecing being done on that front. The point being, of course, that Grand Theft Auto 4’s re-release, if it is indeed on the way, cannot take a similar approach. Looking at what the standard is for a number of high profile remastered releases in the current market, it’s impossible not to have higher expectations from a GTA 4 re-release. Just take a look at something like Ninja Gaiden 2 Black or The Elder Scrolls 4: Oblivion Remastered, both of which are remasters that released this year and ended up impressing many with just how significantly upgraded and overhauled they were over the original experiences in spite of only being remasters on paper. Go back a little further, and look at, say, Metroid Prime Remastered, or Mass Effect Legendary Edition, or The Legend of Zelda: The Wind Waker HD, all of which offered meaningful upgrades and improvements that you didn’t have to squint to even be able to notice. While a Grand Theft Auto 4 re-release doesn’t have to be quite on that same level as far as its upgrades are concerned, with standards and expectations for remasters being higher in general, a low effort port for GTA 4 will be quite a hard pill to swallow by many- as evidenced by the disappointment many felt with the Red Dead port. Rockstar has tried its hand at actual remastered releases in recent years as well, of course, though that sadly didn’t work too well. Grand Theft Auto: The Trilogy – The Definitive Edition launched a handful of years ago on modern platforms and, as most would tell you, did an awful job of remastering and modernizing three gaming masterpieces. Though GTA: The Definitive Edition did definitely put in much more of an effort than the Red Dead port did, the effort that it did put in was lackluster, to put it mildly- and our hope is that Grand Theft Auto 4’s purported release for modern platforms will sidestep those errors as well. The thing with GTA 4 is, of course, that it remains a spectacular game with incredible impressive tech and visuals even today, a ridiculously immersive and authentic urban open world map to lose yourself in, one of the most gripping stories that GTA has ever told, told through who may very well be the series’ best protagonist ever. Grand Theft Auto 4 was an unabashed classic in every sense of the word, so to see it modernized and updated with the right amount of effort and thought put in could be incredibly exciting, not only for returning fans looking to experience Liberty City again, but also for new ones who never played the game when it first came out. Grand Theft Auto 6 will obviously likely up the ante in all those areas by quite some margin when it releases next year, but until then, an enhanced GTA 4 on modern consoles should serve as an excellent appetizer. Here’s hoping that actually does end up happening this time. Note: The views expressed in this article are those of the author and do not necessarily represent the views of, and should not be attributed to, GamingBolt as an organization.

It’s one thing to invent something cool within controlled laboratory environments. It’s entirely another to scale that new baby for sale.  The tension between innovation and commercialization is something we regularly wrestle with at Abstrax. Every morning, we don lab coats and ask the same question: “How do you make money from research done in a lab?”  Balance innovation with commercial reality  Discovery for its own sake isn’t enough. Many R&D-heavy companies discover that brilliant ideas can languish for years if they don’t have a system for bringing them to market.   We decided early on to build that bridge proactively. This meant investing heavily not just in research, but in the systems and machinery that connect lab work to real-world products. Our scientists don’t concoct in ivory towers, they work hand-in-hand with product developers to ensure (most) experiments align with market needs.  This pragmatic approach focuses our curiosity. We poured resources into advanced analytical technology. Among other exquisite toys, we operate an ultra-sensitive aroma analysis platform nicknamed OMNI. With it, we can break down a complex flavor into hundreds of molecular components and pinpoint the ones that matter. By capturing a “3D metabolite fingerprint” of a material (cannabis flower, hop varietals, etc.), we get a complete map of that sample’s aroma chemistry.  Why go to these extremes? Because understanding a flavor at that granular level is the key to replicating it, enhancing it, and ultimately monetizing it. We can identify over 500 distinct compounds in a single hop strain. That level of resolution lets us see opportunities others might miss, like the trace molecules that round out pineapple, or a sulfur compound responsible for “skunky” notes.   From breakthrough to beer glass  To illustrate how lab research turns into revenue, take our recent work with Citra, one of the most celebrated hops in craft brewing. Citra’s appeal lies in its remarkably juicy, complex flavor—think grapefruit, lime, peach, and passionfruit steamrolled into one. This tropical medley makes a Citra-hopped beer delicious. But here’s the rub: Achieving that same flavor consistently at scale is hard. Hops are agricultural products, subject to the whims of weather and yearly variation. The Citra you get this year might not taste exactly like last year’s crop.  This is where our lab-to-market philosophy shines. Using OMNI, we profiled Citra’s chemical makeup in exquisite detail. Armed with that “blueprint,” we developed an Omni Hop Profile extract that mirrors Citra’s flavor profile with uncanny accuracy. For brewers, this is a game-changer. Instead of being at the mercy of Mother Nature, they can rely on our Citra extract to deliver the exact same flavor in every batch, forever. And because it’s made from botanically derived ingredients, it stays true to the clean-label standards brewers abide by. We even worked with veteran brewmasters on pilot brews to fine-tune the extract’s performance in different beer styles. By the time our Citra profile hit the market, it was brewer-tested and production-ready.   Our Citra victory highlights our core principle of reasonable innovation. We didn’t stop at discovering what makes that hop special, we pushed to make it a tangible solution to a real problem. That is the essence of monetizing R&D: moving from “Eureka!” to a viable SKU.   No fluff, only real solutions  In avant-garde industries like cannabis and craft beer, it’s easy to get caught up in hype and bold claims. We prefer a different tack: Let the results speak. If we say our new formulation improves a beer’s shelf-life or an extract boosts an IPA’s aroma, we’ve got the data to back it up. Grounding innovation in evidence keeps us credible and ensures we stay focused on real market value.  We also recognize that not every experiment will pan out, and that’s okay. Part of our system is knowing never to become 100% pot committed. We’ll test 10 ideas, then swiftly double down on the one or two that show commercial promise. By failing fast and smart, we conserve resources for the innovations that count.  The new R&D playbook  Our journey from lab to market hasn’t been quick or easy. It took patience and a willingness to invest up front. But that patience is paying off. Today, Abstrax’s approach is turning niche scientific insights into mainstream products. What others consider to be a cost center is our engine for growth.   When scientists and strategists work in sync, every discovery is viewed through the lens of real-world impact. The healthy tension between invention and commercialization keeps us sharp. As it turns out, the lab and the market are pretty good at balancing each other.   Kevin Koby is CEO and cofounder of Abstrax. 

This post will share how to build an AI journal with the LlamaIndex. We will cover one essential function of this AI journal: asking for advice. We will start with the most basic implementation and iterate from there. We can see significant improvements for this function when we apply design patterns like Agentic Rag and multi-agent workflow.You can find the source code of this AI Journal in my GitHub repo here. And about who I am. Overview of AI Journal I want to build my principles by following Ray Dalio’s practice. An AI journal will help me to self-reflect, track my improvement, and even give me advice. The overall function of such an AI journal looks like this: AI Journal Overview. Image by Author. Today, we will only cover the implementation of the seek-advise flow, which is represented by multiple purple cycles in the above diagram. Simplest Form: LLM with Large Context In the most straightforward implementation, we can pass all the relevant content into the context and attach the question we want to ask. We can do that in Llamaindex with a few lines of code. import pymupdf from llama_index.llms.openai import OpenAI path_to_pdf_book = './path/to/pdf/book.pdf' def load_book_content(): text = "" with pymupdf.open(path_to_pdf_book) as pdf: for page in pdf: text += str(page.get_text().encode("utf8", errors='ignore')) return text system_prompt_template = """You are an AI assistant that provides thoughtful, practical, and *deeply personalized* suggestions by combining: - The user's personal profile and principles - Insights retrieved from *Principles* by Ray Dalio Book Content: ``` {book_content} ``` User profile: ``` {user_profile} ``` User's question: ``` {user_question} ``` """ def get_system_prompt(book_content: str, user_profile: str, user_question: str): system_prompt = system_prompt_template.format( book_content=book_content, user_profile=user_profile, user_question=user_question ) return system_prompt def chat(): llm = get_openai_llm() user_profile = input(">>Tell me about yourself: ") user_question = input(">>What do you want to ask: ") user_profile = user_profile.strip() book_content = load_book_summary() response = llm.complete(prompt=get_system_prompt(book_content, user_profile, user_question)) return response This approach has downsides: Low Precision: Loading all the book context might prompt LLM to lose focus on the user’s question. High Cost: Sending over significant-sized content in every LLM call means high cost and poor performance. With this approach, if you pass the whole content of Ray Dalio’s Principles book, responses to questions like “How to handle stress?” become very general. Such responses without relating to my question made me feel that the AI was not listening to me. Even though it covers many important concepts like embracing reality, the 5-step process to get what you want, and being radically open-minded. I like the advice I got to be more targeted to the question I raised. Let’s see how we can improve it with RAG. Enhanced Form: Agentic RAG So, what is Agentic RAG? Agentic RAG is combining dynamic decision-making and data retrieval. In our AI journal, the Agentic RAG flow looks like this: Stages of Agentic Rag. Image by Author Question Evaluation: Poorly framed questions lead to poor query results. The agent will evaluate the user’s query and clarify the questions if the Agent believes it is necessary. Question Re-write: Rewrite the user enquiry to project it to the indexed content in the semantic space. I found these steps essential for improving the precision during the retrieval. Let’s say if your knowledge base is Q/A pair and you are indexing the questions part to search for answers. Rewriting the user’s query statement to a proper question will help you find the most relevant content. Query Vector Index: Many parameters can be tuned when building such an index, including chunk size, overlap, or a different index type. For simplicity, we are using VectorStoreIndex here, which has a default chunking strategy. Filter & Synthetic: Instead of a complex re-ranking process, I explicitly instruct LLM to filter and find relevant content in the prompt. I see LLM picking up the most relevant content, even though sometimes it has a lower similarity score than others. With this Agentic RAG, you can retrieve highly relevant content to the user’s questions, generating more targeted advice. Let’s examine the implementation. With the LlamaIndex SDK, creating and persisting an index in your local directory is straightforward. from llama_index.core import Document, VectorStoreIndex, StorageContext, load_index_from_storage Settings.embed_model = OpenAIEmbedding(api_key="ak-xxxx") PERSISTED_INDEX_PATH = "/path/to/the/directory/persist/index/locally" def create_index(content: str): documents = [Document(text=content)] vector_index = VectorStoreIndex.from_documents(documents) vector_index.storage_context.persist(persist_dir=PERSISTED_INDEX_PATH) def load_index(): storage_context = StorageContext.from_defaults(persist_dir=PERSISTED_INDEX_PATH) index = load_index_from_storage(storage_context) return index Once we have an index, we can create a query engine on top of that. The query engine is a powerful abstraction that allows you to adjust the parameters during the query(e.g., TOP K) and the synthesis behaviour after the content retrieval. In my implementation, I overwrite the response_mode NO_TEXT because the agent will process the book content returned by the function call and synthesize the final result. Having the query engine to synthesize the result before passing it to the agent would be redundant. from llama_index.core.indices.vector_store import VectorIndexRetriever from llama_index.core.query_engine import RetrieverQueryEngine from llama_index.core.response_synthesizers import ResponseMode from llama_index.core import VectorStoreIndex, get_response_synthesizer def _create_query_engine_from_index(index: VectorStoreIndex): # configure retriever retriever = VectorIndexRetriever( index=index, similarity_top_k=TOP_K, ) # return the original content without using LLM to synthesizer. For later evaluation. response_synthesizer = get_response_synthesizer(response_mode=ResponseMode.NO_TEXT) # assemble query engine query_engine = RetrieverQueryEngine( retriever=retriever, response_synthesizer=response_synthesizer ) return query_engine The prompt looks like the following: You are an assistant that helps reframe user questions into clear, concept-driven statements that match the style and topics of Principles by Ray Dalio, and perform look up principle book for relevant content. Background: Principles teaches structured thinking about life and work decisions. The key ideas are: * Radical truth and radical transparency * Decision-making frameworks * Embracing mistakes as learning Task: - Task 1: Clarify the user's question if needed. Ask follow-up questions to ensure you understand the user's intent. - Task 2: Rewrite a user’s question into a statement that would match how Ray Dalio frames ideas in Principles. Use formal, logical, neutral tone. - Task 3: Look up principle book with given re-wrote statements. You should provide at least {REWRITE_FACTOR} rewrote versions. - Task 4: Find the most relevant from the book content as your fina answers. Finally, we can build the agent with those functions defined. def get_principle_rag_agent(): index = load_persisted_index() query_engine = _create_query_engine_from_index(index) def look_up_principle_book(original_question: str, rewrote_statement: List[str]) -> List[str]: result = [] for q in rewrote_statement: response = query_engine.query(q) content = [n.get_content() for n in response.source_nodes] result.extend(content) return result def clarify_question(original_question: str, your_questions_to_user: List[str]) -> str: """ Clarify the user's question if needed. Ask follow-up questions to ensure you understand the user's intent. """ response = "" for q in your_questions_to_user: print(f"Question: {q}") r = input("Response:") response += f"Question: {q}\nResponse: {r}\n" return response tools = [ FunctionTool.from_defaults( fn=look_up_principle_book, name="look_up_principle_book", description="Look up principle book with re-wrote queries. Getting the suggestions from the Principle book by Ray Dalio"), FunctionTool.from_defaults( fn=clarify_question, name="clarify_question", description="Clarify the user's question if needed. Ask follow-up questions to ensure you understand the user's intent.", ) ] agent = FunctionAgent( name="principle_reference_loader", description="You are a helpful agent will based on user's question and look up the most relevant content in principle book.\n", system_prompt=QUESTION_REWRITE_PROMPT, tools=tools, ) return agent rag_agent = get_principle_rag_agent() response = await agent.run(chat_history=chat_history) There are a few observations I had during the implementations: One interesting fact I found is that providing a non-used parameter, original_question , in the function signature helps. I found that when I do not have such a parameter, LLM sometimes does not follow the rewrite instruction and passes the original question in rewrote_statement the parameter. Having original_question parameters somehow emphasizes the rewriting mission to LLM. Different LLMs behave quite differently given the same prompt. I found DeepSeek V3 much more reluctant to trigger function calls than other model providers. This doesn’t necessarily mean it is not usable. If a functional call should be initiated 90% of the time, it should be part of the workflow instead of being registered as a function call. Also, compared to OpenAI’s models, I found Gemini good at citing the source of the book when it synthesizes the results. The more content you load into the context window, the more inference capability the model needs. A smaller model with less inference power is more likely to get lost in the large context provided. However, to complete the seek-advice function, you’ll need multiple Agents working together instead of a single Agent. Let’s talk about how to chain your Agents together into workflows. Final Form: Agent Workflow Before we start, I recommend this article by Anthropic, Building Effective Agents. The one-liner summary of the articles is that you should always prioritise building a workflow instead of a dynamic agent when possible. In LlamaIndex, you can do both. It allows you to create an agent workflow with more automatic routing or a customised workflow with more explicit control of the transition of steps. I will provide an example of both implementations. Workflow Explain. Image by Author. Let’s take a look at how you can build a dynamic workflow. Here is a code example. interviewer = FunctionAgent( name="interviewer", description="Useful agent to clarify user's questions", system_prompt=_intervierw_prompt, can_handoff_to = ["retriver"] tools=tools ) interviewer = FunctionAgent( name="retriever", description="Useful agent to retrive principle book's content.", system_prompt=_retriver_prompt, can_handoff_to = ["advisor"] tools=tools ) advisor = FunctionAgent( name="advisor", description="Useful agent to advise user.", system_prompt=_advisor_prompt, can_handoff_to = [] tools=tools ) workflow = AgentWorkflow( agents=[interviewer, advisor, retriever], root_agent="interviewer", ) handler = await workflow.run(user_msg="How to handle stress?") It is dynamic because the Agent transition is based on the function call of the LLM model. Underlying, LlamaIndex workflow provides agent descriptions as functions for LLM models. When the LLM model triggers such “Agent Function Call”, LlamaIndex will route to your next corresponding agent for the subsequent step processing. Your previous agent’s output has been added to the workflow internal state, and your following agent will pick up the state as part of the context in their call to the LLM model. You also leverage state and memory components to manage the workflow’s internal state or load external data(reference the document here). However, as I have suggested, you can explicitly control the steps in your workflow to gain more control. With LlamaIndex, it can be done by extending the workflow object. For example: class ReferenceRetrivalEvent(Event): question: str class Advice(Event): principles: List[str] profile: dict question: str book_content: str class AdviceWorkFlow(Workflow): def __init__(self, verbose: bool = False, session_id: str = None): state = get_workflow_state(session_id) self.principles = state.load_principle_from_cases() self.profile = state.load_profile() self.verbose = verbose super().__init__(timeout=None, verbose=verbose) @step async def interview(self, ctx: Context, ev: StartEvent) -> ReferenceRetrivalEvent: # Step 1: Interviewer agent asks questions to the user interviewer = get_interviewer_agent() question = await _run_agent(interviewer, question=ev.user_msg, verbose=self.verbose) return ReferenceRetrivalEvent(question=question) @step async def retrieve(self, ctx: Context, ev: ReferenceRetrivalEvent) -> Advice: # Step 2: RAG agent retrieves relevant content from the book rag_agent = get_principle_rag_agent() book_content = await _run_agent(rag_agent, question=ev.question, verbose=self.verbose) return Advice(principles=self.principles, profile=self.profile, question=ev.question, book_content=book_content) @step async def advice(self, ctx: Context, ev: Advice) -> StopEvent: # Step 3: Adviser agent provides advice based on the user's profile, principles, and book content advisor = get_adviser_agent(ev.profile, ev.principles, ev.book_content) advise = await _run_agent(advisor, question=ev.question, verbose=self.verbose) return StopEvent(result=advise) The specific event type’s return controls the workflow’s step transition. For instance, retrieve step returns an Advice event that will trigger the execution of the advice step. You can also leverage the Advice event to pass the necessary information you need. During the implementation, if you are annoyed by having to start over the workflow to debug some steps in the middle, the context object is essential when you want to failover the workflow execution. You can store your state in a serialised format and recover your workflow by unserialising it to a context object. Your workflow will continue executing based on the state instead of starting over. workflow = AgentWorkflow( agents=[interviewer, advisor, retriever], root_agent="interviewer", ) try: handler = w.run() result = await handler except Exception as e: print(f"Error during initial run: {e}") await fail_over() # Optional, serialised and save the contexct for debugging ctx_dict = ctx.to_dict(serializer=JsonSerializer()) json_dump_and_save(ctx_dict) # Resume from the same context ctx_dict = load_failed_dict() restored_ctx = Context.from_dict(workflow, ctx_dict,serializer=JsonSerializer()) handler = w.run(ctx=handler.ctx) result = await handler Summary In this post, we have discussed how to use LlamaIndex to implement an AI journal’s core function. The key learning includes: Using Agentic RAG to leverage LLM capability to dynamically rewrite the original query and synthesis result. Use a Customized Workflow to gain more explicit control over step transitions. Build dynamic agents when necessary. The source code of this AI journal is in my GitHub repo here. I hope you enjoy this article and this small app I built. Cheers! The post How to Build an AI Journal with LlamaIndex appeared first on Towards Data Science.

There’s a reason that Steven Spielberg has the reputation that he has. For decades, Spielberg has been considered one of the best directors in film history, and he’s got the movies to back that assertion up. Opinions are, quite naturally, going to differ on which movies of Spielberg’s are the best. One movie that often winds up near the top of the list is Saving Private Ryan. The film, which follows a group of soldiers in the aftermath of the D-Day invasion as they are given a mission to receive a fellow soldier in France, is considered one of the greatest World War II epics ever made. Here are three reasons you should watch the movie on Peacock this May. Recommended Videos The opening sequence is iconic for a reason Although Saving Private Ryan may or may not be Spielberg’s best film, almost everyone can agree that the opening sequence, which is an extended look at what the beaches of Normandy were like on D-Day, is among the most riveting and harrowing things he’s ever made. That opening sequence is the most important part of the film, a reminder that the men fighting were just regular people and that they were faced with the kinds of horrors that they would carry with them forever. It’s not easy to watch, exactly, but it is among the most awe-inspiring sequences ever put to film, not least because it feels so raw and visceral. It’s anchored by an unusual premise Dreamworks The question of whether it’s even possible to make an anti-war movie has existed for as long as war movies have, but Saving Private Ryan uses its premise to emphasize the pointlessness of all the death it depicts. The soldiers at the center of the mission have been asked to put their lives at risk to rescue a single man, and many of them lose their lives along the way. Because of this, we come to understand that the soldiers on the frontline have almost no control over their own fates. The movie wants you to wonder why so many people should die for a single man and whether any of this was worth the cost. It features a roster of great performers Dreamworks On top of being the first time that Spielberg and Tom Hanks collaborated on a movie, Saving Private Ryan is filled with great actors from top to bottom. Matt Damon’s Ryan feels like the kind of man worth saving and does a lot with a relatively small amount of screen time. Jeremy Davies, Adam Goldberg, and Tom Sizemore are also standouts, as is a young Vin Diesel, who had not yet become an action icon. It’s an indelible cast, expertly chosen so that they don’t look like movie stars but instead like an actual platoon of young soldiers who are fighting for their country. You can watch Saving Private Ryan on Peacock.

Google I/O, Google’s biggest developer conference of the year, is nearly upon us. Scheduled for May 20 to 21 at the Shoreline Amphitheatre in Mountain View, I/O will showcase product announcements from across Google’s portfolio. Expect plenty of news relating to Android, Chrome, Google Search, YouTube, and — of course — Google’s AI-powered chatbot, Gemini. Earlier this week, Google hosted a separate event dedicated to Android updates: The Android Show. During it, the company announced new ways to find lost Android phones and other items, additional device-level features for its Advanced Protection program, security tools to protect against scams and theft, and a new design language called Material 3 Expressive. Here’s what else you can expect. Gemini and AI Image Credits:Thomas Fuller/SOPA Images/LightRocket / Getty Images AI is the tech du jour, and Google, like its rivals, has been investing heavily in it. A shoo-in for I/O is a new addition (or several) to Google’s flagship Gemini family of AI models. Leaks over the past few weeks suggest that an updated Gemini Ultra model is on the way, Gemini Ultra being Google’s top-of-the-line Gemini offering. With this upgraded Gemini Ultra may come a pricier Gemini subscription. Techcrunch event Join us at TechCrunch Sessions: AI Secure your spot for our leading AI industry event with speakers from OpenAI, Anthropic, and Cohere. For a limited time, tickets are just $292 for an entire day of expert talks, workshops, and potent networking. Exhibit at TechCrunch Sessions: AI Secure your spot at TC Sessions: AI and show 1,200+ decision-makers what you’ve built — without the big spend. Available through May 9 or while tables last. Berkeley, CA | June 5 REGISTER NOW Google offers a single premium tier, Gemini Advanced ($20 per month), to unlock additional capabilities in its Gemini chatbot, which is powered by the company’s Gemini models. But Google may soon launch two new plans, Premium Plus and Premium Pro. It’s not yet clear what benefits might be attached and how these plans might be priced relative to Gemini Advanced. Google will almost certainly talk about Astra, its wide-ranging effort to build AI apps and “agents” for real-time, multimodal understanding. Also probably on the agenda is Project Mariner, Google’s AI “agents” that can navigate and take action across the web on a user’s behalf. Folks on X spotted references to “Computer Use” in the code for Google’s AI Studio developer platform, which could well pertain to Mariner. Everything else Going by the official I/O schedule, Google will have plenty to discuss following The Android Show and I/O keynote addresses. The schedule lists sessions dedicated to Chrome and Google Cloud, Google Play (the Android app store), Android development tools, and Gemma, Google’s collection of “open” AI models. Last year, Google unveiled a few AI-themed surprises at I/O, including a set of models fine-tuned for education applications called LearnLM. An upgrade to Google’s viral podcast-generating NotebookLM could be one such surprise. Leaked code reveals a “Video Overviews” tool that presumably would create video summaries, most likely leveraging Google’s Veo 2 video-generating model. Updated: 3:19 p.m. Pacific: Added a list of new Android features announced during The Android Show.

May 16, 20253 min readIn World First, Baby Receives Personalized CRISPR Gene-Editing TreatmentA CRISPR treatment seems to have been effective for a baby’s devastating disease, but it is not clear whether such bespoke therapies can be widely appliedBy Heidi Ledford & Nature magazine KJ Muldoon, a baby born with a genetic disease that affected his ability to metabolize proteins, has become the first person to receive a bespoke CRISPR treatment. Children's Hospital of PhiladelphiaA baby boy with a devastating genetic disease is thriving after becoming the first known person to receive a bespoke, CRISPR therapy-for-one, designed to correct his specific disease-causing mutation.Little KJ Muldoon, now nearly ten months old, is doing well after receiving three doses of a gene-editing treatment to mend a mutation that impaired his body’s ability to process protein, his parents told reporters this week. But it is too soon to use the word “cure”, says Rebecca Ahrens-Nicklas, a pediatrician at Children’s Hospital of Philadelphia in Pennsylvania, and one of Muldoon’s physicians. “This is still really early days,” she says. “We know we have more to learn from him.”To reach this point, an international team of clinicians and researchers in industry and academia, with support from US government funders and regulatory agencies, raced to develop Muldoon’s therapy in a mere six months. Yet, the drug that it developed, described in the New England Journal of MedicineOn supporting science journalismIf you're enjoying this article, consider supporting our award-winning journalism by subscribing. By purchasing a subscription you are helping to ensure the future of impactful stories about the discoveries and ideas shaping our world today.It’s an ambitious approach that researchers hope will inspire others to harness CRISPR to treat ultra-rare genetic diseases. “This truly is the future for all of these gene and cell therapies,” says Arkasubhra Ghosh, who studies gene therapy at Narayana Nethralaya Eye Hospital in Bengaluru, India, and who was not involved in the study. “It’s really exciting.”Early illnessDozens of people have received CRISPR-based therapies for genetic conditions such as sickle-cell anaemia, but those treatments were designed to be used in many people with the same disorder, regardless of the underlying mutations that caused it. By contrast, researchers tailored Muldoon’s therapy to correct a specific genetic sequence in his genome.Muldoon had inherited two mutations, one from each parent, that meant that he did not produce the normal form of a crucial enzyme called carbamoyl phosphate synthetase 1 (CPS-1). This compromised his ability to process the nitrogen-containing compounds produced when the body breaks down protein. As a result, his blood had high levels of ammonia, a compound that is particularly toxic to the brain.The best treatment for CPS-1 deficiency is a liver transplant, but it would be months before Muldoon became eligible. Meanwhile, each day brought added risk of brain damage or death: only about half of babies with severe CPS-1 deficiency survive long enough to receive a transplant.Ahrens-Nicklas decided to offer the family another option. She and her colleagues had been working with a CRISPR-based technique called base editing, which can make targeted, single-letter changes to DNA sequences. The team was developing ways to quickly and safely tailor a base-editing therapy to correct an individual’s particular mutations. Perhaps now it was time to try the approach in humans, she thought.With the approval of Muldoon’s parents, the researchers enlisted a lengthy roster of collaborators. The team quickly screened for the best base-editing approach and tested it in mice and monkeys. Companies donated proprietary expertise and components. The US Food and Drug Administration fast-tracked its evaluation of the treatment.Rapid deploymentIn just six months, Muldoon received his first dose — a “remarkable” achievement, says Waseem Qasim, a pediatrician at the University College London Great Ormond Street Institute of Child Health, who has used base editing to engineer immune cells to fight cancer.After that initial dose, Muldoon could safely eat the amount of protein recommended for his age, but still needed medications to keep his ammonia levels in check. With a second round of the therapy, the researchers were able to reduce the amount of medicines needed, but could not eliminate his need to take them.Muldoon has since received a third and final dose. His clinicians are carefully reducing his medication dosage, little by little, says Ahrens-Niklas.It’s unclear how this approach could be expanded to treat others with ultra-rare diseases: even when designed to treat hundreds of people, gene therapies and gene-editing therapies are notoriously expensive. “There’s no great answer to this,” says Qasim.For now, each milestone that Muldoon reaches is a tiny miracle to his parents. Earlier this week, his mother, Nicole, walked into his hospital room to find him sitting up by himself in his crib. “We never thought this was going to happen,” she says.This article is reproduced with permission and was first published on May 15, 2025.

Author(s): Adam BEN KHALIFA Originally published on Towards AI. Why are agentic workflows essential for achieving AGI Let me ask you this, what if the path to truly smart and effective AI , the kind we call AGI, isn’t just about building one colossal, all-knowing brain? What if the real breakthrough lies not in making our models only smarter, but in making them also capable of acting, adapting, and evolving? Well, LLMs continue to amaze us day after day, but the road to AGI demands more than raw intellect. It requires agency. Getting Our Terms Straight: AGI, Agency, and Agentic Workflows Before we dive in, let’s define the main concepts here: AGI — Artificial General Intelligence: You can see it as an AI model that can perform any intellectual task a human can. This means not just understanding language or generating images, but adapting, learning, reasoning, and acting across entirely new domains. Agency: The capacity of an entity to act purposefully in its environment to achieve goals. A rock has no agency; a human planning their day has plenty. For an AI, agency means it’s not just passively responding to prompts but actively pursuing objectives. Simply put, it’s the capacity to pursue goals autonomously through planning, acting, and adapting. Agentic Workflows: If agency is the “what”, agentic workflows are the “how”. These are the dynamic processes and systems an AI uses to exercise its agency. Think beyond a simple input-output model. Agentic workflows involve: Autonomous Goal-Setting & Planning: the AI doesn’t just execute a pre-defined plan, it can formulate goals and strategize how to achieve them. Tool Use & Orchestration: like a skilled craftsperson, it can select, combine, and utilize various “tools” (other AI models, databases, APIs, code execution environments) to get the job done. Memory & Learning: it remembers past actions, learns from successes and failures, and adapts its strategies over time. Adaptation in Dynamic Environments: the real world is messy, an agentic AI can adjust its plan when encountering unexpected obstacles or new information. It’s vital to understand the difference here: An LLM calling a weather API is just tool use. An agentic workflow is when an LLM, tasked with “analyzing market trends for a new product,” autonomously decides to:1) search recent financial news, 2) query a sales database, 3) use a data analysis tool to spot correlations, 4) ask a specialized forecasting model for projections, and then 5) compile a summary report, re-evaluating its approach at each step.It’s like the difference between a single musician playing one note, and a conductor leading an entire orchestra. The Limitation of Isolated Intelligence Let’s consider a human analogy. Imagine a brilliant engineer, a genius in their field. Now, strip away their tools: no computer, no internet for research, no pen and paper for sketching ideas or taking notes, no lab for prototyping, no colleagues to bounce ideas off. Confine them to only their thoughts. How much could they truly achieve? Their raw intellect, however vast, becomes severely handicapped when uncoupled from the ability to interact, experiment, and leverage external resources. This “intelligence in isolation” scenario illustrates a fundamental truth: intelligence doesn’t operate in a vacuum. It thrives on interaction, tool use, and the ability to execute plans in the world. If we want AGI, we can’t just build a disembodied digital brain, we need to build something that can act. Agentic Workflows: AI can act like us, and perhaps even better Humans are masters of adapting their “workflows.” A painter uses different tools and processes than an engineer, who uses different methods than a chef. We intuitively understand context, choose the right approach, and even invent new methods when old ones fail. Agentic workflows aim to achieve similar capabilities: Contextual Flexibility: An Agentic AI could switch between “investigative journalist mode” (querying databases, cross-referencing sources, interviewing) and “creative writer mode” (generating narratives, exploring styles) as needed for a complex task. Learning by Doing (and Re-doing): Human learning is an iterative workflow: observe, hypothesize, experiment, analyze, conclude, refine. Agentic systems can embody this, trying approaches, evaluating outcomes, and improving their strategies. Beyond Monolithic Thought: We don’t store everything in our heads. We use notes, computers, books, and critically, we delegate tasks to others. Agentic AI can similarly leverage external knowledge bases, specialized sub-agents, and computational tools, creating a distributed, more powerful form of intelligence. Thinking About Thinking: Humans possess meta-cognition — the ability to reflect on our own thought processes and adjust them. Agentic workflows, with their capacity for self-monitoring and re-planning, are a foundational step towards AI developing its own form of meta-cognition. Inventing New Ways: Perhaps soon enough, an advanced agentic AI won’t just use existing tools and workflows, but identify the need for entirely new ones and even contribute to their creation. A hallmark of true general intelligence. Not Just Helpful, But Mandatory: Why AGI Needs Agentic Workflows These capabilities aren’t just fancy add-ons. They are arguably essential for anything we’d recognize as AGI: Tackling Complexity: Real-world problems are messy, multifaceted, and rarely solved by a single, linear process. Agentic workflows will allow AI to break down these complex challenges into manageable sub-tasks, orchestrating diverse capabilities. Achieving Scale: Imagine trying to manage global logistics, conduct large-scale scientific research, or personalize education for millions with a single, rigid program. Agentic systems offer the modularity and dynamic coordination needed for such scale. Adaptability and Robustness: What happens when the data changes, a tool fails, or an assumption proves wrong? A static AI might grind to a halt. An agentic AI can adapt, re-plan, find alternative solutions, and continue pursuing its goal. It can handle the unexpected. Resourcefulness: Like our engineer, an AGI needs to be able to identify and use the right “tool” (be it a specific algorithm, dataset, or external service) for the job at hand, rather than trying to be a jack-of-all-trades with a single block massive model. Surpassing Human Adaptability The first step is for AI to achieve a human-like ability to set goals, plan, use tools, and adapt through agentic workflows. But the true promise of AGI lies in surpassing these capabilities: Speed: Learn and adapt at speeds incomprehensible to us, iterating through problem-solving cycles in milliseconds. Scale: Manage and orchestrate operations of immense complexity, juggling thousands of variables and “tools” simultaneously. Novelty: Devise entirely new, perhaps counter-intuitive, workflows and solutions to problems that humans haven’t even conceived of. Self-Improvement of Workflows: An AGI that doesn’t just use workflows but actively refines, optimizes, and even discovers fundamentally new and more efficient ways to achieve its goals. Deeper Meta-Learning: Learning how to learn, plan, and strategize more effectively over time, becoming increasingly more intelligent and capable. Long-Horizon Reasoning: Successfully breaking down and navigating extremely complex, multi-stage goals that unfold over extended periods, adapting robustly along the way. Obviously, this is easier said than done. Building true AGI presents formidable challenges: How do we design systems that can reliably plan in open-ended environments? How can they discover and integrate new tools seamlessly? How does the system learn which part of a long, complex workflow was responsible for success or failure? These are active areas of research, pushing the boundaries of what AI can do. Thankfully we are witnessing more and more breakthroughs everyday, and RL — Reinforcement Learning based approaches are showing great promise (which make sense actually, but that’s for another article). Conclusion: Agency as the Cornerstone of AGI The quest for AGI is more than a race for larger models or faster processing. It’s a quest for intelligence that is versatile, adaptive, and purposeful. Agentic workflows provide the framework for such intelligence, enabling AI to move beyond mere pattern recognition to become an active participant in the problem-solving process. Just as human collective general intelligence emerged not merely from neurons, but from networks of thought, culture, and action — we must build AGI not as a single-block model, but as an AI capable of learning, adapting, and acting. Agency, in this light, isn’t just a feature; it’s the fundamental engine that will drive us towards true Artificial General Intelligence. If you liked this article, make sure to follow for more.And you can find me on: Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor. Published via Towards AI

Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More Unpatched systems are a ticking time bomb. Fifty-seven percent of cyberattack victims acknowledge that available patches would have prevented breaches, yet nearly one-third admit failing to act, compounding the risk. Ponemon research shows organizations now take an alarming average of 43 days to detect cyberattacks, even after a patch is released, up from 36 days the previous year. According to the Verizon 2024 Data Breach Investigations Report, attackers’ ability to exploit vulnerabilities surged by 180% from 2023 to 2024. Chronic firefighting makes manual or partially automated patching overly burdensome, further pushing patching down teams’ priority lists. Relying on manual or partially automated patching systems is considered too time-consuming, further reducing patching to the bottom of a team’s action item list. This is consistent with an Ivanti study that found that the majority (71%) of IT and security professionals think patching is overly complex, cumbersome and time-consuming. When it comes to patching, complacency kills Attackers aggressively exploit legacy Common Vulnerabilities and Exposures (CVEs), often ten or more years old. A sure sign of how effective attackers’ tradecraft is becoming at targeting legacy CVEs is their success with vulnerabilities in some cases, 10-plus years old. A sure sign that attackers are finding new ways to weaponize old vulnerabilities is reflected in the startling stat that 76% of vulnerabilities leveraged by ransomware were reported between 2010 and 2019. The misalignment between IT and security teams compounds delays, with 27% lacking cohesive patch strategies and nearly a quarter disagreeing on patch schedules. One of the unexpected benefits of automating patch management is breaking the impasse between IT and security when it comes to managing the patch workload.    “Typically, on average, an enterprise may patch 90% of desktops within two to four weeks, 80% of Windows servers within six weeks and only 25% of Oracle Databases within six months from patch release date”, writes Gartner in their recent report, “We’re not patching our way out of vulnerability exposure.” The report states that “the cold, hard reality is that no one is out patching threat actors at scale in any size organization, geography or industry vertical.” Ring deployment: proactive defense at scale Every unpatched endpoint or threat surface invites attackers to exploit it. Enterprises are losing the patching race, which motivates attackers even more. In the meantime, patching has become exponentially more challenging for security and IT teams to manage manually. Approximately a decade ago, ring deployment began to rely on Microsoft-dominated networks. Since then, ring deployments have proliferated across on-premise and cloud-based patch and risk management systems. Ring deployment provides a phased, automated strategy, shrinking attacker windows and breach risks. Ring deployment rolls out patches incrementally through carefully controlled stages or “rings:” Test Ring (1%): Core IT teams quickly validate patch stability. Early Adopter Ring (5–10%): A broader internal group confirms real-world compatibility. Production Ring (80–90%): Enterprise-wide rollout after stability is conclusively proven. Ivanti’s recent release of ring deployment is designed to give security teams greater control over when patches will be deployed, to which systems and how each sequence of updates will be managed. By addressing patching issues early, the goal is to minimize risks and reduce and eliminate disruptions. Gartner’s ring deployment strategy escalates patches from internal IT outward, providing continuous validation and dramatically reducing deployment risk. Source: Gartner, “Modernize Windows and Third-Party Application Patching,” p. 6. Ring deployment crushes MTTP, ends reactive patching chaos Relying on outdated vulnerability ratings to lead patch management strategies only increases the risk of a breach as enterprises race to keep up with growing patch backlogs. That’s often when patching becomes cybersecurity’s endless nightmare, with attackers looking to capitalize on the many legacy CVEs that remain unprotected. Gartner’s take in their recent report “Modernize windows and third-party application patching” makes the point brutally clear, showing how traditional patching methods routinely fail to keep pace. In contrast, enterprises embracing ring deployment are getting measurable results. Their research finds ring deployment achieves a “99% patch success within 24 hours for up to 100,000 PCs,” leaving traditional methods far behind. During an interview with VentureBeat, Tony Miller, Ivanti’s VP of enterprise services, emphasized that “Ivanti Neurons for Patch Management and implementing Ring Deployment is an important part of our Customer Zero journey.” He said the company uses many of its own products, which allows for a quick feedback loop and gives developers insight into customers’ pain points. Miller added: “We’ve tested out Ring Deployment internally with a limited group, and we are in the process of rolling it out organization-wide. In our test group, we have benefited from deploying patches based on real-world risk, and ensuring that updates don’t interrupt employee productivity–a significant challenge for any IT organization.” VentureBeat also spoke with Jesse Miller, SVP and director of IT at Southstar Bank, about leveraging Ivanti’s dynamic Vulnerability Risk Rating (VRR), an AI-driven system continuously recalibrated with real-time threat intelligence, live exploit activity, and current attack data. Miller stated clearly: “This is an important change for us and the entire industry. Judging a patch based on its CVSS now is like working in a vacuum. When judging how impactful something can be, you have to take everything from current events, your industry, your environment and more into the equation. Ultimately, we are just making wiser decisions as we are not disregarding CVSS scoring; we are simply adding to it.” Miller also highlighted his team’s prioritization strategy: “We have been able to focus on prioritizing Zero-Day and Priority patches to get out first, as well as anything being exploited live in the wild. Using patch prioritization helps us eliminate our biggest risk first so that we can reduce our attack surface as quickly as possible.” By combining ring deployment and dynamic VRR technology, Ivanti Neurons provides enterprises with structured visual orchestration of incremental patch rollouts. This approach sharply reduces Mean-Time-to-Patch (MTTP), accelerating patches from targeted testing through full deployment and significantly decreasing the exposure windows that attackers exploit. Caption: The Ivanti Neurons interface visually manages deployment rings, success thresholds, patching progress and streamlining operational clarity. Source: Ivanti Neurons Comparing Ivanti Neurons, Microsoft Autopatch, Tanium and ServiceNow: Key strengths and gaps When selecting enterprise patch management solutions, apparent differences emerge among leading providers, including Microsoft Autopatch, Tanium, ServiceNow and Ivanti Neurons. Microsoft Autopatch relies on ring deployment but is restricted to Windows environments, including Microsoft 365 applications. Ivanti Neurons expands on this concept by covering a broader spectrum, including Windows, macOS, Linux and various third-party applications. This enables enterprise-wide patch management for organizations with large-scale, diverse infrastructure. Tanium stands out for its robust endpoint visibility and detailed reporting features, but its infrastructure requirements typically align better with resource-intensive enterprises. Meanwhile, ServiceNow’s strength lies in workflow automation and IT service management integrations. Executing actual patches often demands significant additional customization or third-party integrations. Ivanti Neurons aims to differentiate by integrating dynamic risk assessments, phased ring deployments and automated workflows within a single platform. It directly addresses common enterprise challenges in patch management, including visibility gaps, operational complexity and uncertainty about vulnerability prioritization with real-time risk assessments and intuitive visual dashboards. Caption: Ivanti Neurons provides real-time patch status, vulnerability assessments, and risk exposure metrics, ensuring continuous visibility. Source: Ivanti Neurons Transforming patch management into a strategic advantage Patching alone cannot eliminate vulnerability exposure. Gartner’s analysts continue to stress the necessity of integrating compensating controls, including endpoint protection platforms (EPP), multifactor authentication, and network segmentation to reinforce security beyond basic patching. Combining ring deployment with integrated compensating controls that are part of a broader zero-trust framework ensures security, allows IT teams to shrink exposure windows, and better manage cyber risks. Ivanti’s approach to ring deployment incorporates real-time risk assessments, automated remediation workflows, and built-in threat management, directly aligning patch management with broader business resilience strategies. The design decision to make it part of Neurons for Patch Management delivers the scale enterprises need to improve risk management’s real-time visibility.   Bottom line: Integrating ring deployment with compensating controls and prioritization tools transforms patch management from a reactive burden to a strategic advantage. Daily insights on business use cases with VB Daily If you want to impress your boss, VB Daily has you covered. We give you the inside scoop on what companies are doing with generative AI, from regulatory shifts to practical deployments, so you can share insights for maximum ROI. Read our Privacy Policy Thanks for subscribing. Check out more VB newsletters here. An error occured.

The Ontario Science Centre was shuttered on June 21, 2024, with the Ontario government claiming that a structural roof issue made it unsafe. New documents suggest that the engineers inspecting the roof didn’t agree. Photo by Canmenwalker via Wikimedia Commons New documents, obtained by Canadian Architect through a freedom of information request, appear to indicate that Rimkus—the structural engineers hired to assess the Ontario Science Centre’s roof before its closure—did not support closing the Centre for public safety reasons. The draft versions of the report instead recommend regular, routine maintenance to the roofs over the course of the coming 20 years. When read in conjunction with correspondence preceding the closure obtained by Global News, it seems likely that the engineers were pressured by the government to produce a report that would support the shuttering of the Ontario Science Centre. However, such a recommendation was not supported by their research and expertise. Canadian Architect obtained draft versions of parts of the Rimkus report from March, April, and May of 2024. The report, whose final version is dated June 18, 2024, was commissioned to assess the condition of the Science Centre’s reinforced autoclaved aerated concrete (RAAC) roof panels, an outdated form of construction that can pose problems over time if not properly maintained. It was based on on-site investigations conducted by Rimkus from December 2023 to March 2024. In a key section on RAAC Panel Recommendations, the March and April versions of the report clearly recommend to proceed with repairs to a set of panels identified as “high risk”—less than 2.5% of the Science Centre’s roofs—over the summer. It reads: “It is Rimkus’ recommendation to address all 2024 RAAC panel repair (amber risk) locations as a single project. As such, it is our opinion that the most cost effective strategy would be a consistent repair approach. Panel replacement rather than reinforcement is feasible at all areas, and presented in Table 7.” The government seems to have pushed the engineers to go further, asking how the RAAC risk could be entirely eliminated. The May version adds the sentences: “It is recommended that RAAC panels should be replaced at the time of the next scheduled roof replacement, to completely eliminate the RAAC panel risk. Recommended timelines for complete roof assembly and RAAC panel (where applicable) replacements, and budgetary costs, by roof area are shown in Table 8.” In the March, April, and May versions of the report, there is no provision for declining to do the repairs—simply because roof repairs are a relatively routine, straightforward undertaking. RAAC panels at the Science Centre had been routinely replaced in the past. At the time of the inspection, a half dozen panels were identified as being in critical condition, and were already being replaced while the report was completed. A building permit was in progress to repair a section of the roof over the rainforest room. The cost of the immediate repairs needed would amount to around $500,000—or just over $7 million, if coupled with replacing, repairing, and maintaining larger sections of the Science Centre’s roofs that year. And yet, the language apparently wasn’t strong enough for the Ontario government, whose communications staffers aimed to justify the closure as a public safety issue. For the final version of the report, government officials seem to have pressed the engineers to answer the question: what if we just don’t do the repairs? Wouldn’t the only totally safe option, in that case, be to close the building? As a result, the final version of the report included the following passage: “Where replacement or reinforcement cannot be completed within the recommended time frame, then one of the following supplemental risk mitigation options is recommended and should be implemented in conjunction with the snow/load monitoring program: Option 1: Restricted access or full closure to prevent any persons from walking in areas where high risk panels are present. Option 2: Installation of temporary shoring (reinforcement) supporting the underside of RAAC panels. Options 3: If shoring is not possible, installation of horizontal hoarding near the underside of hard ceiling levels, or other building interferences (sprinkler mains/process piping etc.) The above three recommended options are listed in order of preference, with option 1 completing [sic] eliminating the risk to public or staff.” As I have written before, while a quick reading of this passage may seem to suggest “full closure” of the Science Centre, a closer look shows that access would only need to be restricted at the areas “where high risk panels are present.” The high-risk panels represent a miniscule portion of the Science Centre’s roof, almost entirely in non-exhibition areas. Texts between Infrastructure Ontario minister Michael Lindsey and a top communications staffer, obtained by Global News, point to the Ministry’s communications with structural engineers Rimkus in the lead-up to the announcement of the Ontario Science Centre’s closure. The texts also show that the Ministry of Infrastructure was in tight communication with the Premier’s Office (PO) about the closure.   Text messages between Infrastructure Ontario CEO Michael Lindsay and the agency’s top communications staffer, from the run-up to the closure announcement, suggest that the agency was working closely with Rimkus and pressing executives to tailor its language before the report was submitted. “The executives at Rimkus look like survivors of the apocalypse. [laugh/cry emoji],” one of the texts reads. The presence of a typo in the report’s added section also suggests a rushed addition. When the final report was sent, a member of Rimkus’s team seems to have also referenced conversations that took place between the two parties. “Please refer to revised final summary report (R2) attached, with updated notes/paragraphs, as discussed,” they wrote on June 18. While some back-and-forth between consultants and public clients is expected, it is unusual that this contact would be with the Ministry’s top brass—particularly for a routine, technical report. Also notable is that while the Rimkus report released to the public by Infrastructure Ontario was signed by the engineers, it was not stamped—an indication that the engineers may not have felt entirely comfortable with how the report landed, and were unwilling to be professionally liable for it. In an email to top Ministry of Infrastructure and Infrastructure Ontario officials after the closure announcement, an Ontario Ministry director of communications shares a few questions that may arise. One is: “Q: Is this [closure] just a convenient excuse to move [the Ontario Science Centre] to Ont Place?” He suggests the reply: “This is a health and safety issue. Full stop.” But for the engineers that actually inspected the roof, closure should never have been on the table: the roof issue could—and should—have been addressed with routine repairs. The post New documents reinforce that Science Centre closure was not supported by engineers appeared first on Canadian Architect.