More than 40% of pull requests on GitHub now start with an agent's first draft. In March it was 17%. This week GitHub made agentic workflows generally available inside Actions. Writing code is no longer the constraint. Review capacity and ownership are leadership decisions now.
Lab829
@lab829.bsky.social
Lab829 is an advisory shop for technology, product and delivery leadership. Senior advice on the hard calls, then help to build what gets decided. lab829.com
A sketch worth keeping doesn't have to stay in the browser. Download Project in CodeRunner hands over the full Vite app, ready for your own editor. Unzip it and open it in your own editor. Free in the browser.
Salesforce now ships an agent kill switch and hard spending caps. Its own announcement admits fewer than half of companies running agents test them on live traffic. A stop button is the easy part. Knowing when to press it takes an owner and a live eval.
Open CodeRunner, then lose the Wi-Fi. The editor still loads and the React preview still updates as you type. Handy on a train, or on hotel Wi-Fi that drops every ten minutes. JavaScript, TypeScript and React, free in the browser.
Code written in CodeRunner already has the shape of a real React project. Files sit where Vite puts them: index.html, public and src, with main.jsx and App.jsx inside. Same shape as a Vite project. Every open file gets an editor tab. Free in the browser.
Change one line in App.jsx and the Live React Preview catches up in about 0.5s. That gap between typing and seeing is where React starts to make sense. CodeRunner runs JavaScript, TypeScript and React, free in the browser.
The bet was that autonomy would come from replacing experience. It turns out autonomy depends on it. Whether an AI-assisted system holds up in production still comes down to judgment. Knowing which output looks right but isn't.
Enterprise trust in autonomous AI agents is falling, and the data says it should. 61% of organizations deploying agents had a customer-facing failure after a feature passed internal evals (VB Intelligence). A human review gate is how trust gets built.
Agent adoption in banking will be decided by permission design, not model quality. JPMorgan lets outside agents read balances but not pay bills with a customer's login. Amex gives the agent a scoped credential, never the card number.
For banks, blocking outside AI agents is the easy half. The hard half is what they offer the agent instead. Our Wealth Finder prototype gave Claude ten narrow tools on synthetic data. Reads were open. Writes stayed drafts until the user approved them.
A demo proves an idea works. A product has to work every day, at a cost someone approved. 21 checks for the gap between the two: ownership, quality, data, cost, security, change, support. Tick what's true today. lab829.com/pilot-to-pro...
AI Pilot-to-Production Checklist | Lab829
What usually breaks between an AI demo and a live product. A checklist across ownership, quality, data, cost, security, change and support, with why.
lab829.com
Most AI pilots that stall don't have a model problem. They have an ownership, data or governance problem the demo hid. 10 yes/no questions to find yours. No sign-up, answers stay in your browser. lab829.com/ai-readiness...
AI Readiness Scorecard for Product Teams | Lab829
Ten yes/no questions across people, product, data, governance and delivery. Score your team and see where AI work usually stalls before it ships.
lab829.com
Aloe Theme is here! 🚀 And it's all about good vibes and calm greens while coding 🧘 Most dark themes are loud. Aloe isn’t. A calm green VS Code theme built for focus and WCAG-friendly contrast. → marketplace.visualstudio.com/items?itemNa... #vscodetips #a11y #codeeditor #visualstudiocode
Aloe Theme - Visual Studio Marketplace
Extension for Visual Studio Code - Aloe is a dark color theme for Visual Studio Code. Inspired by the Aloe plant.
marketplace.visualstudio.com