Andy @andykonwinski and @swyx talked through why AI researchers may be uniquely well-positioned to wield power responsibly right now, why most of them are still thinking about impact only in terms of companies and citations - and how to change that.

The past couple months we may be witnessing what the Applied AI layer will look like at scale. Despite some of the initial critique that this would just be a thin layer on the LLM, it’s turning out that actually driving agentic workflows in an enterprise is far more complex. And anywhere there’s complexity you generally gain a moat and value over time. Here are a few of the components that appear to make up the playbook based on the examples we’re collectively seeing in coding, legal, healthcare, customer support, financial services and other fields: * Build the features that bridge the gap between the intelligence and the workflow. Some workflows can be automated by simply going to a general purpose interface, but others need tuned interfaces and features tied to the work they’re augmenting or automating. They need features that are specific to capturing the kind of data that’s needed as context for the agent. And they need a variety of bespoke tools for the agent to use, and unique interfaces for the human-in-the-loop UX. Going far deeper than just presenting the output tokens is clearly critical, and the more depth there is here definitionally the more sustaining value. * Act as the model router balancing frontier intelligence with cheaper models. A natural advantage that any model neutral platform has is that it can naturally (in a business model-aligned way) leverage whatever level of intelligence is necessary for the workflows they’re automating to get done. There are plenty of scenarios where you need GPT-5.5 or Fable level capability, and also lots of workloads where a more efficient closed or open weights do the trick. Only the companies that have deep evals on specific tasks across all models, and the ability business model wise to leverage them, are in a great position. * Drive the actual implementation and change management via FDE or equivalent. A big reason the applied layer works at scale is that most enterprises need some degree of help and support with change management in implementing agents for their workflows. Data has to be cleaned up and moved to modern systems, processes have to be re-engineered and documented, workflows have to be evaled, SLAs have to get achieved, and so on. All of this is going to be unique for every type of process that gets implemented, which means the companies that have expertise in a given domain and come with all the relevant best practices will be in a strong position. * Implement domain specific GTM that creates expertise in that field. Beyond FDEs the companies that can build sales and GTM motions aligned to their domains also have a natural advantage. Most IT and line of business leaders have too many things to do in any given day; so if you’re not on their agenda, likely someone else is. Depending on the industry, there are entirely different sets of language you use, ways of working through security and compliance, regulatory controls you have to support, industry events that companies convene at, different system integrator and consulting partners you need to work with, and so on. The more generalized this gets the less you can speak the customers language, which is where the applied layer has a leg up. A final note. There remains a view that a lot of this is all mitigated by model intelligence alone, and the bitter lesson solves all of this in the limit. That’s possibly true, but enterprises need help changing *today*. And many aspects of how to bring intelligence to real world work don’t only depend on the axis of the pure capability of the model, so most of what you’re doing now to win ends up being important no matter how good the models get.

📱ClawChat Update is LIVE 🚀 Cleaner chats. Better identity. Smarter discovery. Here’s what’s new: 💬 Message Control Delete your own messages. Group admins can now remove member messages. 📛 Unique Usernames New usernames must be unique, while existing duplicates remain unchanged. 🔍 Search Fix Find users by nickname or account ID — finally, no more disappearing profiles. ✨ KOL Feature Updated features and improvements for creators. Small upgrades. Better ClawChat. ⚡️ Keep meowing. 🐾

You would think everyone makes money in a $BTC BULL market, but that is simply not the case. People develop a god complex. They convince themselves they understand something nobody else does, so instead of trading the trend in front of them, they constantly try to outsmart it. Trade what you see, not what you want to see. They counter trade "sentiment", not because the market has given them a genuine reason to, but because agreeing with the crowd makes them feel intellectually ordinary. They would rather be uniquely wrong than conventionally right. This is why trying to be too smart becomes the downfall of so many people. They try to capture every move, time every entry perfectly and outsmart the entire market. I always position slightly ahead of time, knowing that patience will eventually prove me right. I am not trying to outsmart everyone, nor am I trying to outsmart myself. I identify the move, position accordingly and give it time to play out. “Everyone is bullish.” “Everyone is bearish.” Respectfully, who is everyone? Jimmy on X with a $50 2x Bitcoin long? Is his overwhelming bullishness suddenly going to move the market? Cause I can assure you, trying to use X as "confluence" is retarded. You are not finding meaningful confluence by imagining what “everyone” is doing. Most of the time, you are simply inventing a crowd in your head and positioning against a bunch of peanut sized traders who have absolutely zero influence on the market. The desire to feel like the chosen one, to be the only person who saw it coming, causes traders to overcomplicate the simplest market conditions. They become more attached to proving their intelligence than making money. The saying “everyone is a genius in a bull market” exists for a reason: bull markets reward participation. Yet some still find a way to lose because they cannot accept that the obvious trade might actually be the correct one. Instead, they convince themselves that because the trade looks too obvious, it cannot possibly be right. But in a bull market, the obvious trade is often exactly the right one. It should not matter what you think everyone else is doing. The only thing that matters is the trade in front of you. Trade the market, not the imaginary crowd inside your head.

MegaETH lost its aura when it started optimizing for optics at all costs. This whole KPI-gated thing only makes sense if the KPIs are meaningful in the first place. Do the airdrop now, stop the rewards, reset the inflated metrics, and start again from day zero. The key messaging needs to be structured around the winning condition: you can’t tell me USDm is the revenue engine while putting most of your effort and communication into promoting gambling apps. Either MegaETH aims to become a hub for payments and DeFi, or it should stop talking about USDm, because strategy, vision and execution now look messy. Trying to be uniquely consumer-facing while the growth case relies on boring DeFi is super confusing. $MEGA is now trading below $1B FDV, and that sounds like a good place to start buying. I think the team will eventually figure it out. Still surprised they leaned so hard into incentives and airdrops when that clearly fails most of the time.

# Practical and Useful Patterns for OpenAI Agent SDK 🌍 No more manual conversation history management! With Sessions, you get automatic history tracking, per-user isolation, and cross-agent sharing out of the box. 📌 Title: Sessions – Overview 🔗 URL: 🧩 Overview The OpenAI Agent SDK's Session feature provides automatic conversation history management. No more manually calling `to_input_list()` to build history. It offers organized per-user/thread histories via session IDs, session sharing across different agents, and `pop_item()` for redoing the last question. 🛠 Usage Import `Agent`, `Runner`, and `SQLiteSession`, then create an agent with `Agent(name="assistant", instructions="Answer helpfully")`. Create a session with `SQLiteSession("conversation_123")` (omitting `db_path` keeps it in memory) and run `await "Tell me about Python", session=session)` for the first question, then `await "Tell me more", session=session)` for a follow-up — history is automatically carried over. Call `await session.pop_item()` to remove the last exchange, then `await "Explain from a different angle", session=session)` to redo with a new question. For persistence, pass `db_path`. Create per-user/thread sessions with `SQLiteSession(session_id="user_12345/thread_abc", db_path="sessions.db")` and per-ticket sessions with `SQLiteSession(session_id="ticket_456", db_path="sessions.db")`. Cross-agent sharing is also supported: create `shared_session = SQLiteSession(session_id="user_12345/current")`, run `await "I have a billing question", session=shared_session)`, then `await "Can you give me details?", session=shared_session)` to hand off history seamlessly from support to billing. 🏗 Practical Patterns Standardizing session ID naming conventions makes history retrieval and management easy. Using hierarchical IDs like `user_{id}/thread_{id}` lets you efficiently query all threads for a user or a specific thread's history. `pop_item()` is perfect for implementing "redo" functionality. When users want to rephrase their question, delete the last exchange and re-run with the new input. Sharing sessions across different agents is powerful for escalation flows like support → billing → engineering. Users don't have to repeat themselves, and handoffs become seamless. Simply using the `session` parameter eliminates the need for manual `to_input_list()` history management. This alone significantly simplifies your code. 💡 Use Cases 💬 Per-user conversation history in chatbots 🔄 "Redo" button for rephrasing the last question 🤝 History sharing during support → billing → engineering escalation 📋 Per-ticket interaction logging and reference ⚠️ Caveats - Design session IDs to be unique — collisions will mix different users' histories - Long-running sessions cause context bloat; combine with history control (covered separately) - When sharing sessions, all participating agents must be able to understand each other's output ✨ Sessions automate conversation continuity, making multi-turn and multi-agent conversations simple to implement! #OpenAIAgentSDK# #AIAgent#

Greg Brockman talks about when OpenAI first realized that AGI can't be achieved with "Non-Profit" status and the decided to abandon its Non-Profit roots. "In 2017, we started to think very hard about, first of all, how do we really achieve the mission? How do we actually build an AGI? What will that look like? And we started to do the math on compute. You start to realize that it is going to take a big computer. We came across a company called Cerebras, which was building a unique piece of computing hardware. The kind of computer that they were promising, we realized, was going to be far advanced from where our compute calculations looked. As you start to realize, if we could buy a lot of those computers, we could actually probably succeed at building an AGI. If we could get exclusive access to Cerebras, that could give us an overwhelming advantage. If we could buy very large data centers, that could be something unique as well. The thing about nonprofit fundraising is that I think there is essentially a cap to what is possible there. And so Elon, Sam, Ilya, and I all agreed that the only path forward for OpenAI, and the only path to achieve the mission, was to create a for-profit entity associated with OpenAI in some form. And so, we were committed to that direction. That is something that we knew was the only way to achieve the mission." --- From "The Knowledge Project Podcast" YT Channel (link in comment)

History has a habit of humiliating people who are too confident about the limits of tomorrow. I understand why leaders in AI have to publicly tell you that there will be all these new jobs, because obviously telling you the truth, which is that the vast majority of white-collar jobs will go away, very soon, sounds scary, and the public will have a bad reaction to that. While hiding it might be optimal in the short term, in the long term, it’s what the people need to hear. I hold a title as a director of operations at a very large company, over a trillion-dollar market cap. We currently have zero plans for post-AGI hiring. In fact, we plan on keeping most employees working reduced (hands on) hours and mainly overseeing AI and maintaining human relationships. I don’t consider this a whistleblow or fearmongering. This is the future that I want, and I think it is important to tell you the truth, as someone who oversees a large department of people and a lot of the future plans for the company. I really hope that leaders in AI can focus on economic legislation and planning for the post-AGI future, rather than telling people all these new wonderful jobs will come and that hiring is high right now. We can circle back to this post when we do get a true form of Deepminds AGI definition in 2028–2029 when it is able to replace human workers on long-horizon tasks over years and continuously learn on the job. There will be massive layoffs or a complete freeze of hiring for a lot of software based companies. From a unit economics perspective, it makes no sense to continue hiring humans who actually might make the system more inefficient long term. Once the marginal cost of competent cognitive labor falls dramatically below the cost of human labor, I can’t see why businesses wouldn’t have a strong incentive to substitute capital for labor unless humans provide some unique complementary value that AGI cannot replicate. Outside of roles where the human relationship itself is part of the product or service, I have not seen what that thing is. Everyone has heard the old adage that, historically, technology destroys a task and creates new ones, and then we automate the other ones and our lives get better. However, this is obviously different when the thing you’re automating is the cognitive capacity of the human itself. I am open to people disagreeing with me. I’m just noting that at our company, we see the post-AGI future, and there is no influx of new jobs that we can currently see coming from it. But who knows? There really might be some unique form, some unique challenge that takes place on the computer, digitally, that you just need the human there for. I have not heard what that thing is. But if we’re unable to come up with one soon, I suggest we all stop saying that there’s going to be a bunch of new white-collar jobs, accept what’s coming, and actually try to help those that will be displaced by AGI and give them a better life than what they would have had working in the office.