Read all three parts of @dexhorthy's series.
In part 3 he stops arguing on vibes and runs a benchmark.
Most teams still tune their AI dev setup by feel. If you don't have a benchmark, you can't hill climb.
@acatovicx I agree that it will take a while so your take on the gap is still relevant.
I find that I trust Elons progress on this more than most robotics companies though for some reason. It is needed for his ambitious Space plans so no stopping until its done. π
@acatovicx Thats what Elon sees as well. π If he gets his way Optimi will solve that lag in the physical world. Working in refineries and factories to bring them up to speed.
i had a lot of fun chatting with @mattpocockuk today about how i was able to land 2,500 PRs last month! Matt is a wonderful interviewer so i think the interview turned out really interesting
both of our skill plugins work great together, so i recommend giving both a try and picking the best skills that suit your workflow
https://t.co/RS0Mxgoy8n
It doesnt make sense to review every line anymore. Same as you donβt look at all the datapoints when training an ML model. You need to set up solid verification and validation. Agents collecting proof of the feature being secure/compliant, live up to spec and doesnβt introduce any regressions. Encode what good looks like and measure results.
New city, new conference! π
Just had a great session at CTO Craft Amsterdam during the last few days and now in Copenhagen for goto;
Speaking on Friday about the growing gap between development velocity and trust. And how you can set up closed loop verification to successfully ship fast with high confidence.
I'm heading to Amsterdam today! π³π±
Tomorrow I'm speaking on stage at CTO Craft Con about guardrails and verification in the agentic loop.
Every developer is shipping more independently, and agents are still bad at following intent. You can't expect each contributor to reinvent the building blocks, architectural constraints and design principles that keep a product cohesive and aligned with the business.
The way to keep that alignment is to encode what good looks like into your SDLC: use platforms for consistency across the org, and codify your team's tacit knowledge as guardrails and context.
See you there! π