Bobby

@artillain @endpointarena At least I can log in!

@bobbyjoins @endpointarena Oh, you'll definitely be having more fun with Microsoft tech support

>i got banned from OpenAI for "Cyber Abuse" >no idea what I did >paste the ban notice into Codex >ask it to figure out what triggered the ban >Codex found that I asked it for an API key to my own server >Codex writes appeal >Codex submits appeal >a few minutes later appeal auto-approved by some AI at OpenAI banned by AI, convicted by AI, defended by AI, and pardoned by AI in about 10 minutes

@brian_armstrong I get the point and agree with saving intelligently. But this doesn't show what he says. It's not growing exponentially. It stopped 6 months ago. They didn't curb cost till they stopped growing exponentially. Different initiative and premise

How to keep AI spend flat while token usage grows exponentially: Not with friction and spend alerts. With better defaults, routing, and caching. Better Defaults (not Usage Caps) – Engineers can choose any model they want, but defaults matter. We’re experimenting with defaulting to open weight models like GLM 5.2 and Kimi 2.7 through our LLM gateway, while still encouraging engineers to choose the right model for the task. 91% of our employees were never hitting their usage caps, so instead of lowering caps and driving up alerts, we're moving to cheaper defaults. Note that code reviews use a diversity of models, so they can check each other's work. Better Routing – In our custom harnesses, we preprocess prompts and route to the best model for the job, considering cache hits and model pricing. For instance, you may want a frontier model for planning, but not for execution where they can be overkill. Ultimately, humans shouldn't be choosing models - AI can automate this task. Better Caching – Cache misses are the easiest way to drive your cost up. All of our requests are cache aware, so we’re reusing a warm cache wherever possible. For example, our cache hit rate went from 5% → 60% in LibreChat once properly implemented. Keep Context Lean – Start fresh sessions when switching tasks. Scope file context narrowly. Disconnect unused tools. Don't just compact. The goal isn't fewer tokens used, it's fewer tokens wasted. Better Visibility – Our engineers can use as many tokens as they want, from whatever model they want, but we’ve made usage visible – and the more you spend on AI, the more impact we expect. The goal isn't to suppress usage. It's to build the infrastructure that makes exponential growth sustainable. Putting this into practice has cut our AI spend nearly in half, while our token usage continues to grow.

There is a use case for ai kanban boards ... if you wanted a fixed price compute setup then you need strict prioritization. This is why humans use it. You have a team of X people and they need to prioritize what to do now vs later etc. But if it's just for dag and you're trying to have the agents crank at max speed it's not an ideal design.

kanban doesn’t make sense for coding agents we tried it 6 months ago. every task just ended up in the “needs review” column

@0xmcat I really don't even know what he's saying. Judgement wasn't the only difference between the two types. And two people with an llm don't arrive at the same judgement decisions.

Perfect example of someone who does not maintain software and likely never has. Awful engineers can cause more damage today than ever, and to some degree because of this mindset. Being a 10x engineer means so much more than holding a unique ability to traverse solution spaces. These engineers coordinate, critique, and uncover things that don’t fit into a 1M context window. They have agency, creativity, curiosity, and drive to close feedback loops and solve problems that others don’t even see. The philosophy of “we don’t need distinguished engineers because we have junior ones with LLMs” is flipped. A 10x engineer equipped with LLMs can accomplish things that a team of 10 mids never will- and increasingly so. If you think that the next trillion dollar software company will be built by anything other than the top echelon of software engineers; I couldn’t disagree more.

@rohanpaul_ai I don't understand this at all. Being a 10x engineer wasn't only about judgement. Also a 10x engineer is absolutely more effective with AI than a 1x. There are points to be made about the delta collapsing, but the rest of the commentary sounds like nonsense.

Chamath on how AI agents are making the "10x engineer" distinction disappear because the most efficient "code paths" are now obvious to everyone. Just as AI solved chess and removed the mystery of the best move, AI is doing the same for coding, making the process reductive and removing technical differentiation. "I'm going to say something controversial: I don't think developers anymore have good judgment. Developers get to the answer, or they don't get to the answer, and that's what agents have done. The 10x engineer used to have better judgment than the 1x engineer, but by making everybody a 10x engineer, you're taking judgment away. You're taking code paths that are now obvious and making them available to everybody. It's effectively like what happened in chess: an AI created a solver so everybody understood the most efficient path in every single spot to do the most EV-positive (expected value positive) thing. Coding is very similar in that way; you can reduce it and view it very reductively, so there is no differentiation in code." --- From @theallinpod YT channel (link in comment)

@AndrewCurran_ It's like a tad funny when Claude code started the Anthropic rise and it was a side quest. I'm not sure what to take from that but the irony is present

The WSJ is reporting that OpenAI is about to take a hard turn into enterprise.

@jeffrey_way Twitter will also tell you the number of agents and size of workflow and length of processing are the goal not any real outcome. "I implemented x and it improved my business by $y" is a real story.... and that's hard.

I'm pretty convinced that very very few developers are actually doing these super-sophisticated workflows that involve a dozen agents working concurrently in their own respective worktrees. Twitter/X would have us believe the opposite is true, weirdly.

@VibeMarketer_ Most businesses are messy and hard to run. Launch sales agent sounds great, but improve sales by 15% is harder. The implementation layer of software is now agents and agents are more intertwined with the messiness of business than software was before.

when anthropic ships something like this, pay attention to what they're NOT building. they're not building your custom sales agent.. integrating with your CRM or designing workflows for your ops team. they're building the platform. the implementation is up for grabs.

@BeatZap_io @jarvis_best Or fridge becomes iot device

@jarvis_best They treating AI integration like when Bluetooth speakers came out and everything was like a toothbrush plus a speaker or shoes and a speaker.

Don’t save for retirement. It won’t matter. The singularity is here. AI lamp.

@HedgieMarkets @mjesuele But when the metric is tokens and ai use not value this is what you get

🦔I suspect the answer is agentic workflows running unsupervised with massive context windows, spinning up subagents and validating things that don't need validation. There's a difference between using AI as a tool and letting AI orchestrate itself. The second one burns tokens exponentially.

🦔 Jason Calacanis says his company hit $300/day per agent using Claude's API at only 10-20% capacity, which scales to around $100,000/year per agent. Chamath Palihapitiya added that he's now asking "what's the token budget for our best devs?" and said AI-assisted developers need to be at least 2x as productive just to justify the cost. He said this is actively happening inside his company or he'll run out of money. My Take This was always the obvious trajectory. AI providers subsidized usage to drive adoption, and now the subsidies are ending. The consumer plans are likely loss leaders subsidized by VC money, and the gap between what individuals pay and what it actually costs to run these models is closing fast. I'm struck by the surprise from people who should know better. These are sophisticated tech investors just now realizing that running agents 24/7 burns through tokens at rates that dwarf human salaries. A human engineer runs on coffee, remembers context from years ago, builds institutional knowledge, and doesn't rack up exponential costs the longer they think about a problem. Agents waste tokens constantly, researching and validating things that don't need validation, spinning up subagents for simple tasks when a straightforward approach would work fine. The companies that fired engineers to replace them with AI agents are learning that you can't negotiate with an API bill the way you can renegotiate a salary. And unlike employees who might stick around during a rough patch, the meter just keeps running. Hedgie🤗

@thdxr We optimize what we measure.

lines of code has become a good metric now i so often see simple projects that are noticeably way more LoC than i'd expect great sign of slop

@arscontexta I don't disagree with the point but we have arrived at any new trend just post a "<trend> graph" article and you can do numbers

@ashleymayer More common in writing the writer and the thinker are the same person. With coding that's not always the case

Question for my technical friends: I'm a big believer that writing is thinking. It's why I'm hesitant to outsource any writing that matters (like an investment memo) to LLMs, slop factor aside. Is coding thinking? And by that I mean, if you fully outsource the coding work and only prompt and give feedback, do you lose anything? Do you begin to think about problems differently, less creatively? Or is all the thinking in the scoping and the actual coding was always a tax?

@Hesamation To be fair, making business decisions off bad data is more common than not. Just usually from a person not ai

if it takes a whole company 3 months to realize AI has been making up critical metrics and nobody ever checked once, I doubt AI is the stupid one here. also maybe not use 3B local models to run a whole company.

I'm not commenting on how efficient it actually makes people or rollout. I'm saying people make these claims about the impact that don't make sense mathematically. Everyone has an agenda. If developers were running 20 agents simultaneously and not needing to review PRs open ai would be merging in like 60 PRs per person per day. It's a ludicrous claim.

@bobbyjoins @lennysan @sherwinwu This is fair to an extent as we stand right now. Though we are seeing the models shipping faster, output is up, Open AI are achieving that 70% with metrics around quality etc. By the end of the year that 70% will be far higher. Rollout beyond that takes time

My biggest takeaways from @sherwinwu: 1. AI is writing virtually all code at OpenAI. 95% of the engineers use Codex, and engineers who embrace these tools open 70% more pull requests than their peers, and that gap is widening over time. 2. The role of a software engineer is shifting from writing code to managing fleets of AI agents. Many engineers now run 10 to 20 parallel Codex threads, steering and reviewing rather than writing code themselves. 3. The average PR code review time has dropped from 10-15 minutes per PR to 2-3 minutes. Every pull request at OpenAI is now reviewed by Codex before human eyes see it, and Codex surfaces suggestions and catches issues up front. This allows engineers to focus on more creative and strategic work while dramatically increasing productivity. 4. The models will eat your scaffolding for breakfast. When building AI products, don’t optimize for today’s model capabilities. The field is evolving so rapidly that the scaffolding (vector stores, agent frameworks, etc.) that seems essential today may be obsolete tomorrow as models improve. 5. Build for where the models are going, not where they are today. The most successful AI startups build products that work at 80% capability now, knowing the next model release will push them over the line. 6. Top performers become disproportionately more productive with AI tools. AI tools amplify the productivity of high-agency individuals, so the gap between top performers and everyone else is widening. The ROI on unblocking and empowering your best people compounds faster than ever in an AI-augmented environment. 7. Most enterprise AI deployments have negative ROI because they’re top-down mandates without bottom-up adoption. Success requires both executive buy-in and grassroots enthusiasm. Sherwin recommends creating a “tiger team” of technically-minded enthusiasts (often not engineers) who can explore capabilities, apply AI to specific workflows, and create excitement throughout the organization. 8. The one-person billion-dollar startup is coming, but with unexpected second-order effects. As AI makes individuals more productive, we’ll see not just billion-dollar solo founders but an explosion of small businesses: hundreds of $100M startups and tens of thousands of $10M startups. This will transform the startup ecosystem and venture capital landscape. 9. Business process automation is an underrated AI opportunity. While Silicon Valley focuses on knowledge work, most of the economy runs on repeatable business processes with standard operating procedures. There’s massive potential to apply AI to these workflows, which are often overlooked by the tech community. 10. The next two to three years will be the most exciting in tech history. After a relatively quiet period from 2015 to 2020, we’re now in an unprecedented era of innovation. Sherwin encourages everyone to engage with AI tools and not take this moment for granted, as the pace of change will eventually slow. 11. AI models will soon handle multi-hour tasks coherently. Today’s models are optimized for tasks that take minutes, but within 12 to 18 months we’ll see models that can work on complex tasks for upward of six hours. This will enable entirely new categories of products and workflows. 12. Audio is the next frontier for multimodal AI. While coding and text get most of the attention, audio is hugely underrated in business settings. Improvements in speech-to-speech models over the next 6 to 12 months will unlock significant new capabilities for business communication and operations.

@shakoistsLog Great point ... even if not +EV it wasn't far from zero. The negative impact of doing your job poorly is probably scaling at the same rate as the positive impact of doing it well.

dead weight used to be marginally +EV cause you'd just assign them to manage cloud configs or somesuch. but now slop cannons are actively scary.

@ItakGol They changed everything

Suddenly it hit me. What happened to DeepSeek? Sora? GitHub Copilot? Mail0? Llama? Cursor? MS Copilot? Mistral? Perplexity? What happened?

@nathan_covey This has always been unsettling to me. Because it shows there is a disconnect between how it processes and reality ... and I don't know where else those disconnects may exist in less obvious ways.

A funny thing about coding with AI is that it still thinks in human timelines. AI: "It will take 1-2 weeks to fully integrate this new design into your app. Migrate gradually; no need to do everything at once." Me: "No. Do the whole thing right now." *does it in 15 minutes*