Profile avatar
lawrencejones.dev
Engineer at https://incident.io/. Previously @GoCardless | Writes at http://blog.lawrencejones.dev | @lawrjones on Twitter
278 posts 616 followers 359 following
Regular Contributor
Active Commenter

The amount of Lime bike green I can see in the London naked cycle photos is making me reconsider ebiking as a social good

Was fun changing our Old Street billboard to focus on eng hiring last Friday. Went a few rounds with the team and felt this represented us well!

Wrote up a journey of optimising an LLM prompt for latency, reducing the time from 11s to just 2s by being careful with tokens. Loads of lessons here for people working with AI! incident.io/building-wit...

Yesterday we launched a microsite of stories from our AI team. To my knowledge, it’s the best repository of genuinely useful AI engineering content on the internet to date. I’m so proud of the team for this and can’t thank our designers enough for how amazing it looks. incident.io/building-wit...

It's been great fun sharing our approach to AI with Gergely, born out of the last year hitting our heads against AI. This is an awesome article that I'd encourage anyone building systems with generative AI to read! Thank you for the feature!

8 years ago Chris and I dealt with a horrible Postgres incident where RAID array battery failure was an extremely painful red herring. Good to see he’s back in the game after all that time. These disks must really be something!

Sometimes I wonder how many burritos a week is too many burritos

How is this a thing

I have a package that’s been stuck in customs for a week. Is there anything more irritating than a fedex tracker giving you zero info and saying “ETA Tuesday 11th 11am” when it’s now the next week 😂

Crazy to me that people can say this when ChatGPT/Claude/etc have hundreds of millions of daily active users. I have not once ever had a use for a blockchain payment. All my colleagues use AI daily. What is making people so resistant to seeing this?

Was playing with some CI build time improvements yesterday and booting big old machines is something I will always enjoy. 96 cores go brrrrrr

The average AI product experience is bad right now, and it’s because teams haven’t figured it out how to build with these tools yet. Building a prototype is 10% of the effort. The rest of the journey is uncharted waters and can be a real struggle. blog.lawrencejones.dev/ai-mvp/

Gotten into the habit of getting AI to check my code piece-by-piece as I build systems. Doesn't replace a human or a review, but it frequently catches issues I'd otherwise miss or only find once I'd deployed. Can't recommend this enough!

This was a fun chat about how we're building an AI incident responder, which: - Checks your Grafana dashboards - Looks at recent PRs - Finds past incidents like this one And drops a message just like this one I got this morning when I broke our evals 😂