I've turned the way I build with AI agents into a kit. If you've never written code, it takes you from download to a working app in your browser, built the way a senior engineer would. If you're an engineer, it's the same system without the hand-holding.
@JorgeCastilloPr it's surprising the ones doing the interviews aren't suffering with code atrophy as well if they're using AI. I can't imagine trying to interview or test someone now, I'd just say 'You're absolutely right'.
Everyone says AI is taking software engineers jobs.
Mate, you make spreadsheets and send emails. ChatGPT web UI can do that, never mind a suite of agents.
@alexanderOpalic depends what you're shipping. I don't have a line by line memory like I once could the "thats in that file around this function", but I know the structure, architecture and conventions. We don't enforce or allow slop, just don't have to do as much steering any more
Half my feed says AI is going to kill SaaS, the other half says that's nonsense.
Yes, you can knock up a basic copy of any app these days - but who's maintaining it, evolving it, securing it?
Knock up 6 tools internally and you'll be killing yourself managing them.
If you've built something with AI and shared the link, spend ten minutes checking who else can open it. I wrote about a report that 35,578 of 37,571 live Grok-built apps opened with no sign-in. It's not one tool. It's what happens when publishing is one click away.
Google is changing the Gemini app so chats you have with History turned off are kept for around two months, up from 72 hours. Off now means hidden from you, not deleted. If you're pasting client code, contracts or API keys into a consumer chatbot, assume they're still on someone's server in January.
@ThePrimeagen Launch day is the worst day to judge it. OpenAI's Codex lead said the early-access folks were testing a build from before a lot of last-minute changes, so half the takes are about a model that doesn't exist any more. A week of real tickets will tell you more.
@emollick Finding vulnerabilities got cheap almost overnight, but patching still moves at change-board speed. The bottleneck is now testing and rollout, so the teams with decent staged deploys and quick rollbacks are the ones who'll sleep this month.
@theo Per-token pricing tells you almost nothing about long-running work. A model that's cheaper per token but needs three retries and a babysitter costs more per merged PR, and cost per finished task is the number I'd love a bench to report.
OpenAI says GPT-6.2 Sol scores within half a point of GPT-6.2 Astra on FrontierCode 2.0, a test of whether code changes are good enough to merge, at about a fifth of the cost. If your team runs the biggest model on every ticket, you're paying five times over for a rounding error.
@buildwtim I'm not sure it's a case of a complete replacement, more of a reduction in labour and resources and a renaming/positioning of what a developer does. Someone still needs to steer the ship so to speak
You don't need to know what a terminal is to start. Step one in the kit is "pick your tool": Cursor, Claude or Codex. Paste one setup prompt, and about 30 minutes later a starter app is running in your browser. Eight cards take it from there. £39:
https://t.co/2blbOmOedC
@0xAX It really comes down to the spec and planning up front. Finding the edge cases to begin with in the tasks to complete. Then at the end reviewing against that. I don't particularly read every line, but I know exactly what needed to be done and will review against that
All week my feed's been 3 day prep kits, 90 day prep kits, and a Simpsons clip "predicting" a blackout on the 17th.
Then the night before, someone hijacks a website tool and 178 sites, the BBC and HSBC included, start serving code that nicks card details.
Coincidence from the hackers... or all part of the plan 👀
Maybe people should've heeded the advice to withdraw cash after all.