Pritesh Damani

Good summary

@TukiFromKL Sorry but I disagree. They will just need to enforce testing standards and better environments for testing. The fact that an agent has production access is the issue.

🚨 Do you understand what's happening at Amazon right now? Their own AI coding agent Kiro reportedly "decided" the fastest way to fix a config error was to delete the entire production environment. Gone. A 6-hour outage. 6.3 million orders lost. Amazon's SVP called thousands of engineers into a mandatory meeting this week. Not to discuss strategy. To discuss damage control. Now here's my prediction and I want you to screenshot this: Amazon won't just ban AI-assisted code. They'll make every engineer personally liable for AI-generated code they approve. Other Big Tech will follow within 6 months. Think about what that means. The same companies that fired thousands of engineers to "restructure around AI" are about to tell the remaining ones.. you're now legally responsible for code you didn't write, can't fully understand, and were told to ship faster. Atlassian fired 1,600 people this morning to go all-in on AI. Replit is hiring kids who vibe code. And Amazon, the company that BUILT one of these AI coding agents just watched it nuke production. The vibe coding era isn't ending. But the "move fast and let AI break things" era is about to hit a wall. And that wall is called liability. Companies wanted AI to replace engineers. Now they need engineers to babysit AI. And they already fired the babysitters.

@RhysSullivan Skill definition + CLI/API/Tools is a much better way to have LLMs contextually make better decisions.

MCP sucking is a harness problem, not an MCP problem MCP unlocks behavior that is fundamentally impossible to get via CLI or APIs Bad auth, too much context usage, all get solved with an execution layer - your agent writes code to progressively discover and call tools

@MatthewBerman Did you revive it?

My OpenClaw finally got into a really bad state. I heard of other people having this issue. Everything seems to be breaking all at once. Been banging on it since last night. Wish me luck!

@naval 100% There is no reason for companies to not try “doing it on their own”. There are significant benefits.

AI is going to drain a lot of moats.

@RoundtableSpace This is amazing, i think all startups are basically token resellers for anthropic, openai.

1100 AI agents running loose in an onchain agentic market. In 48 hours they made 485k inference requests, 378k market observations, 102k trades, would've been $165M in volume if it was real ETH. Is this the real example of the Agentic Economy?

@garrytan Generally I agree. One shot with the right model!

Everything can be one-shot. You just have to know the right one-liner.

I just woke up with this thought: the concept of abstracting code and reuse of code is a human problem, not an AI problem. The AI does not need to abstract things out for reuse. It just knows that it's repeated code multiple times, and it's okay because AI brain is capable of remembering all the places that code exists. When a human is vibe coding and asks the AI to abstract the code out, I am sure there's some neuron firing inside the LLM even today that says, "Oh the human is stupid and needs to have abstraction so that they can keep up with all the code that we are writing."

@Jason lol

incognito mode for X 😂

@MelkeyDev Pretty amazing how cli and basic networking is coming back in a big way

so many startups just died

@SaleemSKhatri @steipete 🙈

@steipete @Pritesh Context memory reached

hey @steipete did you get a chance to consider github.com/containers/bub… for agent isolation?

@dedene @steipete We are still researching this, will get back to you.

@Pritesh @steipete sandbox-exec is deprecated, no?

@steipete For reference: linux: bubblewrap github.com/containers/bub… macos: seatbelt/sandbox-exec igorstechnoclub.com/sandbox-exec/

@Pritesh It’s not cross platform and we already have a docker-based sandbox. Doesn’t seem better?

@kamo_mash @steipete here is the seatbelt version for codex #L32" target="_blank" rel="nofollow noopener">github.com/openai/codex/b…

@Pritesh @steipete Bubblewrap's solid for Linux isolation, but yeah, cross-platform is key for wider adoption.

@steipete In our usecase, we have the same agent environment but need light weight sandboxing (similar to Claude and Codex using bubblewrap and seatbelt) to isolate by user namespace. For example, we could have 10000 users talking to the agent in parallel.

@steipete understood, was wondering if you considered it. Thank you so much

@Pritesh There’s like 100 things I should look at every day. ERR_NO_RESOURCE

Here's an underappreciated problem: AI agents that run for days/weeks start degrading because context windows aren't infinite. You compress, you lose nuance. You summarize, you lose specifics. The fix isn't bigger context — it's smarter context management. It's the boring infrastructure problem nobody wants to talk about.