En 30 secondes
Theo Browne describes changing his approach after an AI model failed to identify or fix a problem. He used the agent to locate relevant code and build a browser-console tool, while bringing his own observations and hypotheses to the investigation.
- 1Change the questionWhen the model could not find or fix the problem, Theo shifted to testing multiple hypotheses.
- 2Build a testing toolHe asked for changes that could be tried in production through console commands and observed immediately.
- 3Divide the workThe agent found code and built tools quickly; Theo contributed the information and theories that guided the work.
À retenir
- An AI model may help investigate a problem even when it cannot identify the cause itself.
- Theo asked the agent to build a console-accessible tool for testing changes quickly.
- The agent accelerated code discovery and experimentation while Theo supplied the observations and theories.
Lire le texte remanié, chapitre par chapitre
Using the model to investigate
Theo describes moving from asking a model to solve a problem to using it to test theories. A custom browser tool lets him change options and observe the effects.
At this point I realized the model was not going to solve the problem for me, or even find it. So I changed my mindset: instead of waiting for an answer, I thought about how to use the model to test as many different debugging theories as efficiently as possible.
I asked it to make changes easy to experiment with in production, ideally through console commands so I could see an immediate difference. It wrote a window binding that created __T3GPU, giving us a custom tool for switching options back and forth while investigating.
The agent could find things in the codebase faster than I could and build custom tools to test my theories better than I could. But I was still the one bringing the actual information to the investigation.
Sur la même chaîne
38:47EN
Theo - t3․gg· 28 oct. 2026Theo’s Long-Running Agent Experiments: Verification, Compute and CostTheo Browne · Ping LabsA skateboard game and two compiler projects show how repeated checks and flexible compute can sustain agent work—and how expensive those loops can become.
36:07ENTheo - t3․gg· 27 oct. 2026Theo on Token Costs, Coding Agents, and the Price of AI CapabilityTheo Browne · Ping LabsA viral estimate of Lauren's AI spending prompts a closer look at cache pricing, benchmarks, and the changing work of coding agents.
30:37ENTheo - t3․gg· 26 oct. 2026Theo’s Three Skills for Reviewing and Shipping with Coding AgentsTheo Browne · Ping LabsHow sol-loop, full-send and full-stop encode review, verification and different levels of merge authority.
0:43ENTheo - t3․gg· 23 oct. 2026Theo Distinguishes Security Vulnerabilities from Active AttacksTheo Browne · Ping LabsA short explanation of why disclosed software flaws and reported attacks are different kinds of security news.
29:22ENTheo - t3․gg· 22 oct. 2026Theo on Security Vulnerabilities, Stolen API Keys and AITheo Browne · Ping LabsA look at the gap between vulnerability disclosure and patching, and a speculative account of why stolen inference keys may be valuable.
31:13ENTheo - t3․gg· 20 oct. 2026Theo on Why Smarter Models Help More Than DevelopersTheo Browne · Ping LabsFrom game modifications to compiler experiments, Theo revisits his assumptions about who benefits from more capable AI.