Vinny Carpenter

Code is exhaust. The bottleneck moved upstream: context, intent, judgment, and ownership. The 2nd edition of The Bottleneck Is Never the Stack is out. vinny.dev/book #AgenticAI #EngineeringLeadership

Jev doesn't write. You hand it a situation and typed questions, and it returns structured decisions with probabilities. The work becomes defining those decisions and measuring how often they're wrong, since a fast answer can feel certain and still miss. vinny.dev/blog/2026-10-03-je…

GSD Task Manager launches on EarlyHound today. I've spent a year building it around the Eisenhower matrix. It's free, works offline, and doesn't need an account. If it's ever kept something out of your Do First pile, a review there would make my day. earlyhound.com/products/gsd-…

Launching StillView on @macpiphq on Nov 25 🚀 A native Mac image viewer for your photos, with on-device AI and OCR. Come say hi: macpip.com/apps/stillview

GSD Task Manager launches on @macpiphq on Nov 25 🚀 GSD ("Get Stuff Done") is a privacy-first. Have a look before the day: macpip.com/apps/gsd-task-man…

An AI refactor came back clean, tested, and aimed at the wrong problem. The model solved the problem I wrote down, not the one we had. As agents get better at execution, the bottleneck moves to framing. Models climb frames. Leaders build them. vinny.dev/blog/2026-05-24-th…

If you don't have access to Claude Duet, try the link below: claude.ai/duet-invite/xYHMu9…

I asked my six AI agents for self-reviews. Each one described its own work well, and none seemed to know it had five coworkers. When two teams own the same job, better self-reviews won't fix it. You fix the org chart. vinny.dev/blog/2026-11-06-i-…

#PlatformEngineering #SoftwareArchitecture #EngineeringLeadership #AgenticSDLC 7/7

New essay: Architecture Is What Agents Can't Refactor vinny.dev/blog/2026-11-10-Ar… Prefer to listen? The audio edition runs 22 minutes. What's the last change your team shipped that a revert couldn't undo? 6/7

Blackout windows, audit evidence, and sign-offs paced everything after that, and a faster coding agent wouldn't have moved a single date. So I think architecture reviews should stop debating code structure, and architects should sit where the agreements are. 5/7

I watched those costs set the calendar on a data center migration program I led this year. Our teams cut over 932 applications in one weekend. 4/7

What stayed expensive is whatever a revert can't undo: • live data • API contracts other teams have built on • identity boundaries • the org chart 3/7

That's the test I landed on after looking at what agents have done to large migrations. Airbnb pointed an LLM pipeline at about 3,500 test files and converted 75% of them in four hours. Code structure used to be the hard-to-change part of a system. It's now the cheap part. 2/7

Could your team revert this change tomorrow without asking anyone? If the answer is yes, it's design, and agents can churn it as often as you like. If the answer is no, it's architecture. 1/7

I merged 64 agent-built PRs across three apps before lunch, and none of them reached users by lunch. The bottleneck moved to acceptance: deciding whether the evidence is good enough to ship. I'll delegate the typing, not the accountability. vinny.dev/blog/2026-11-04-64…

Frontier labs asking to slow down AI probably mean it. That doesn't decide who benefits. Bruce Yandle's Bootleggers and Baptists theory explains why: a fixed compliance cost is a rounding error for a big lab and a wall for a 40-person startup. vinny.dev/blog/2026-09-14-bo…

Omarchy finally put Linux on my desk because someone made hundreds of small setup decisions for me. Good defaults beat better Linux. The ThinkPad install took two minutes, and most of that was me hunting for the Secure Boot switch. vinny.dev/blog/2026-09-18-om…

The easiest way to make AI look profitable is to leave half the costs off the ledger. Count context prep, review, rework, escaped defects, and governance next to the subscription. Then measure value per verified outcome instead of tokens generated. vinny.dev/blog/2026-06-08-th…

Your platform's defaults teach engineers how work gets done, and agents learn the same lessons. If break-glass is the only path for routine cleanup, people will use it for routine cleanup. Build the missing pipeline instead of sending another reminder. vinny.dev/blog/2026-09-22-yo…

Last Saturday I merged 64 pull requests between 9:30 and 11:38 a.m. Eleven agents worked in two waves under one lead agent. They built across all three GSD Task Manager clients: • 27 PRs for web • 22 for Apple • 15 for Android 1/4