Yeah pretty much. Have you seen Polsia and its ilk? Maybe "trivial" would be too strong a word but... in 2026 it's not hard.
That's my point. You couldn't tell an unemployed farm worker to go start their own farm. They probably don't have the land or substantial capital it takes. But an unemployed software engineer just doesn't need anything like that to go into a business built on AI.
With farming, you couldn't just start your own farm, because it requires farmland, and there's only so much of that. But those 6 software engineers can start their own companies, fire up their own team of agents. There's no limit to how many companies can exist in the world.
I recently started having my AI assistant help clean up my email gradually. (Using stumpy.ai for what it's worth.)
The way I do it is every morning we go through recent emails in my inbox one at a time. If I want to mark it as spam, delete it, add it to my calendar, whatever, I explain to the agent why in detail. Over time it builds up an understanding of how I handle a lot of things, it needs to show me less and less, and it handles more and more on its own.
I also told the assistant to check my email on its own once per hour and auto-action what it can. That helps keep junk from building up, and it alerts me via SMS if something high priority shows up (e.g. user reporting a bug).
Point is there was never a point where it just ran for a long time and magically cleaned everything up just how I'd have wanted. I have like 7k emails in my inbox, that wouldn't be practical. But the number is going down now gradually, instead of up. I've had a chance to teach it and let it establish trust that it's doing things the right way. Which feels safer.
Agreed. People aren’t ready for this, even (maybe especially) on HN.
Everyone’s hung up on how nobody really does waterfall. Or course. But a LOT of people are vibing their code and making PRs and then getting buried in code reviews. Just like the article says, you can’t keep up that way. Obviously. Only agents can review code as fast as agents write it. But I find as of recently that agents review code better than people now, just like how they write it better. Gotta lean into it!
> They don't know that the reason you price things the way you do is rooted in a competitive dynamic that never got written down anywhere.
So maybe you should write it down?
I see this going differently than they do. An exoskeleton that makes you 5% stronger is not a game changer. Companies that lean fully into agents won’t just be “humans, but a little better.” They’ll move orders of magnitude faster, make decisions in less time, do it ask more efficiently. It will be no contest.
Depending on what you mean by claw-like, stumpy.ai is close. But it’s more security focused. Starts with “what can we let it do safely” instead of giving something shell access and then trying to lock it down after the fact.
For one thing they were just early. Whatever measurements people made of AI six months ago are invalid. It’s a different animal now.
Plus you get a wildly different payoff the more you can take humans completely out of the loop. If it writes the code but humans review, you’re still bottleneck. If it designs and codes and reviews and goes back to designing, and so on, there’s no effective speed limit.
Big businesses aren’t going to work that way though. Which is why we shouldn’t be looking to them as thought leaders right now.
Author here. I've been a programmer for 25 years. Elixir, C, Ruby, PHP, Python, FoxPro. About a month ago I stopped writing code entirely and switched to designing through conversation with AI agents.
The 60x number is real but I know it'll be controversial. It's lines of code in a month vs. what I'd produce in a year by hand. Your mileage will vary. I'm not claiming everyone gets this. I'm saying the range of individual experiences is so wide that averages are meaningless.
Yes, this is another post about vibe coding. But it's a real product with real users at this point, more than a weekend project, and I think the health and communication effects are worth talking about even if the productivity claim doesn't land for you.
Along the same lines: I want the context window percentage visible at all times, not just when it drops below 10%. By that point it's too late to do anything useful. I can't even get it to finish up and dump its state to a file before the window is full. If I could see the percentage the whole time, I could pace my work and wrap things up cleanly instead of slamming into the wall.
I'm building a platform of AI agents, and each agent can have its own email address. AgentMail handles that. You create an inbox via their REST API and they POST to your webhook when mail arrives.
On the agent side, it just gets tools: send email, reply to email, list inbox, read message. Those tools call the AgentMail API. So the fake implements the same interface.. same send/reply/list/read methods, but recording calls instead of making HTTP requests. You can pre-populate inboxes with test messages, simulate "username taken" errors, etc.
AgentMail is actually one of the simpler fakes because there's no internal state to maintain. Sending a message doesn't affect what you'd read back from your own inbox. Some of the other fakes (like the database or file storage) need to actually simulate state in memory so writes are visible to subsequent reads. This one is closer to a stub.
> The Digital Twin Universe is our answer: behavioral clones of the third-party services our software depends on. We built twins of Okta, Jira, Slack, Google Docs, Google Drive, and Google Sheets, replicating their APIs, edge cases, and observable behaviors.
Came to the same conclusion. I have an integration heavy codebase and it could hardly test anything if tests weren't allowed to call external services. So there are fake implementations of every API it touches: Anthropic, Gemini, Sprites, Brave, Slack, AgentMail, Notion, on and on and on. 22 fakes and climbing. Why not? They're essentially free to generate, it's just tokens.
I didn't go as far as recreating the UI of these services, though, as the article seems to be implying based on those screenshots. Just the APIs.
But, is that different from how we already work with humans? Typically we don't let people commit whatever code they want just because they're human. It's more than just code reviews. We have design reviews, sometimes people pair program, there are unit tests and end-to-end tests and all kinds of tests, then code review, continuous integration, Q&A. We have systems to watch prod for errors or user complaints or cost/performance problems. We have this whole toolkit of process and techniques to try to get reliable programs out of what you must admit are unreliable programmers.
The question isn't whether agentic coders are perfect. Actually it isn't even whether they're better than humans. It's whether they're a net positive contribution. If you turn them loose in that kind of system, surrounded by checks and balances, does the system tend to accumulate bugs or remove them? Does it converge on high or low quality?
I think the answer as of Opus 4.5 or so is that they're a slight net positive and it converges on quality. You can set up the system and kind of supervise from a distance and they keep things under control. They tend to do the right thing. I think that's what they're saying in this article.
Building it now. Basically raw agents you can talk to over any channel like Slack/Telegram/etc. (Should have SMS and voice calling working shortly.) Can connect to your email/calendar. Files and sqlite for memory/storage. Optional sandbox for coding or whatever. It's all a bit rough but working.
Had a similar thought since I started using the Slack MCP in Claude Code. It's handy for instance during an incident to be researching the problem, digging through Sentry or Clickhouse or the code and have it post updates directly to our #engineering channel for the team to see. But... they can't reply. Or rather they can but Claude has to poll each thread or channel to see replies which is a pretty clumsy workflow.
So anyway long story short I made something like Clawdbot but in the cloud: https://stumpy.ai/
Didn't occur to me to design it to run locally and leave running on my machine. You can't close your laptop or Clawdbot dies? It can read all your files? Rather run agents in the cloud. I gave them sandboxes (Fly sprites) so you can still have them do software development or whatever.
> Your job is to deliver code you have proven to work.
First of all, no it’s not. Your job is to help the company succeed. If you write code that works but doesn’t help the company succeed, you failed. People do this all the time. Resume padding, for example.
Sometimes it’s better for the business to have two sloppy PRs than a single perfect one. You should be able to deliver that way when the situation demands.
Second, no one is out there proving anything. Like formal software correctness proofs? Yeah nobody does that. We use a variety of techniques like testing and code review to try to avoid shipping bugs, but there’s always a trade off between quality and speed/cost. You’re never actually 100% certain software works. You can buy more nines but they get expensive. We find bugs in 20+ year old software.