Skip to main content
The Rework Podcast

AI Revisited

35 min episode · 2 min read
·

Episode

35 min

Read time

2 min

Topics

Productivity, Design & UX, Artificial Intelligence

AI-Generated Summary

Key Takeaways

  • Agentic AI vs Auto-complete: AI agents running autonomously in terminal mode from scripts represent a fundamental shift from intrusive auto-complete tools in code editors. These agents execute bash commands, run programming tasks, and search the web independently for 10 seconds to multiple minutes, producing work developers can keep rather than interfering with their typing flow and thought process.
  • HackerOne Security Review Automation: 37signals deployed AI to pre-screen security vulnerability reports from external researchers, filtering low-quality submissions from legitimate threats. The system accesses historical report data and researcher credibility scores to identify high-value reports requiring human attention, functioning like spam detection applied to security research quality rather than blocking malicious content outright.
  • Console Access Log Auditing: Biweekly reviews of programmer access to production systems and customer data get automated through AI analysis. The system verifies that staff members access data only within granted customer permissions, handling the tedious work of checking 42-plus access events per review cycle that previously required manual human verification to ensure compliance.
  • 80-20 Code Generation Rule: AI agents consistently deliver 80 percent complete solutions that developers can either iterate with the agent or finish manually. The remaining 20 percent requires human refinement, but even 20 percent completion provides value if the output contains no counterproductive errors requiring more cleanup time than writing from scratch would take.
  • Multi-Model Draft Strategy: Using OpenCode terminal interface to query Claude, Gemini, OpenAI, and open-weight models simultaneously on the same problem generates five different working solutions in 2.5 to 6 minutes each. This approach surfaces diverse implementation ideas and working prototypes faster than single-model iteration, even when the final solution combines elements rather than accepting any single draft.

What It Covers

David Heinemeier Hansson explains 37signals' shift toward AI adoption after years of hesitation. The breakthrough came from agentic AI models running autonomously in terminal environments rather than auto-complete tools, plus dramatic improvements in model capability during late 2025. He details internal applications for security review automation and debugging assistance.

Key Questions Answered

  • Agentic AI vs Auto-complete: AI agents running autonomously in terminal mode from scripts represent a fundamental shift from intrusive auto-complete tools in code editors. These agents execute bash commands, run programming tasks, and search the web independently for 10 seconds to multiple minutes, producing work developers can keep rather than interfering with their typing flow and thought process.
  • HackerOne Security Review Automation: 37signals deployed AI to pre-screen security vulnerability reports from external researchers, filtering low-quality submissions from legitimate threats. The system accesses historical report data and researcher credibility scores to identify high-value reports requiring human attention, functioning like spam detection applied to security research quality rather than blocking malicious content outright.
  • Console Access Log Auditing: Biweekly reviews of programmer access to production systems and customer data get automated through AI analysis. The system verifies that staff members access data only within granted customer permissions, handling the tedious work of checking 42-plus access events per review cycle that previously required manual human verification to ensure compliance.
  • 80-20 Code Generation Rule: AI agents consistently deliver 80 percent complete solutions that developers can either iterate with the agent or finish manually. The remaining 20 percent requires human refinement, but even 20 percent completion provides value if the output contains no counterproductive errors requiring more cleanup time than writing from scratch would take.
  • Multi-Model Draft Strategy: Using OpenCode terminal interface to query Claude, Gemini, OpenAI, and open-weight models simultaneously on the same problem generates five different working solutions in 2.5 to 6 minutes each. This approach surfaces diverse implementation ideas and working prototypes faster than single-model iteration, even when the final solution combines elements rather than accepting any single draft.

Notable Moment

David describes an AI agent using a C debugger to diagnose a Rails console bug, working through multiple failed hypotheses before identifying the exact problematic commit and generating a working patch. He acknowledges he theoretically possessed the skills to solve this but would not have invested the hours required, choosing instead to work around the issue indefinitely.

Know someone who'd find this useful?

Episode Transcript

Welcome to Rework, a podcast by thirty seven Signals about the better way to work and run your business. I'm Kimberly Rhodes joined this week by David Heinemeier Hanssen, cofounder and CTO of thirty seven Signals. Well, we have gotten a lot of questions over the years about AI and incorporating AI into our products. We've typically said, like, we're just not quite there yet. We're seeing how things go, but I feel like there's been a little bit of a shift recently internally, at least, from announcements that I've seen. So I thought we would talk a little bit about that today. David, why don't you just jump right in? I know we've kind of not been hesitant. I don't know if hesitant is the right word, but it seems like we're, like, jumping in a a bit more now. Yeah. I think, for me, AI really changed in the later half of 2025. The models got better as they've been getting better for a while, but we gave them more powers, especially when it comes to development. There's a new or not even that new, but approach of instead of having the AI be in your editor, auto completing things, trying to help you write something that you're in the process of writing, they moved out of that and into the terminal running in agent mode, which basically just means that they're running on their own from a script, from a plan where you tell them to do something, and they go off using the tools available on the computer. They use your terminal. They run bash commands. They might even run some programming. They start looking things up on the web. They've been doing that for a while. But when you combine these things that they have control over your computer, which sounds a little scary, but when you see it in action as a agent doing development alongside you, not sort of on top of you, it really changes the game. Not just in how it feels, which to me was a really important part of it. I played around a fair bit with the auto completion paradigm of AI as it was for developers in the initial phase, and I didn't like it at all. I did not like this idea that I have a thought in my head coming out through my keys and then boom, the AI tries to did you think of this? Did you think of that? Can you shut up for a second and just let me type the damn thought out of my head? Now I know there are others who rather enjoyed that and saw some great benefit out of it. I didn't like it. And that was where a lot of my opposition to using AI as a developer tool was coming from that I could see the advantages. I could see the promise, at least I should say. But it didn't feel like it was in …

Get the full transcript (6,359 words) + summary by email — free

One-time email with the complete transcript and AI summary of this episode. No account needed.

One email, no spam. We’ll also show you what SignalCast does.

Browse all The Rework Podcast transcripts →

You just read a 3-minute summary of a 32-minute episode.

Get The Rework Podcast summarized like this every Monday — plus up to 2 more podcasts, free.

Pick Your Podcasts — Free

Keep Reading

Books, tools, and gear mentioned in this episode

SignalCast may earn commission on purchases via these links.

Tools

  • OpenCodeRecommended
    Using OpenCode terminal interface to query Claude, Gemini, OpenAI, and open-weight models simultaneously on the same problem generates five different working solutions in 2.5 to 6 minutes each.
  • ClaudeRecommended

    by Anthropic

    Using OpenCode terminal interface to query Claude, Gemini, OpenAI, and open-weight models simultaneously on the same problem generates five different working solutions in 2.5 to 6 minutes each.
  • GeminiRecommended

    by Google

    Using OpenCode terminal interface to query Claude, Gemini, OpenAI, and open-weight models simultaneously on the same problem generates five different working solutions in 2.5 to 6 minutes each.
  • OpenAIRecommended

    by OpenAI

    Using OpenCode terminal interface to query Claude, Gemini, OpenAI, and open-weight models simultaneously on the same problem generates five different working solutions in 2.5 to 6 minutes each.

company

  • David Heinemeier Hansson explains 37signals' shift toward AI adoption after years of hesitation.
  • 37signals deployed AI to pre-screen security vulnerability reports from external researchers through HackerOne, filtering low-quality submissions from legitimate threats.

More from The Rework Podcast

We summarize every new episode. Want them in your inbox?

Similar Episodes

Related episodes from other podcasts

Explore Related Topics

This podcast is featured in Best Business Podcasts (2026) — ranked and reviewed with AI summaries.

Read this week's AI & Machine Learning Podcast Insights — cross-podcast analysis updated weekly.

You're clearly into The Rework Podcast.

Every Monday, we deliver AI summaries of the latest episodes from The Rework Podcast and 192+ other podcasts. Free for one show.

Start My Monday Digest

No credit card · Unsubscribe anytime