Justin Crites

Builder at @DreambaseAI

[Justin Crites @dreambase_jc]

@supabase @DreambaseAI Huge thanks to @laktek, Matt Johnston, and the Supabase team for the invite and the lightning fast answers. We're prepping to be ready for production on day one.

Full write-up with every number: https://t.co/P4X7z7DTUU

[Justin Crites @dreambase_jc]

The new @supabase Compute is awesome. We moved our @DreambaseAI data agent onto it during the private alpha, and answers came back 1.7x faster at 4.7x lower model cost.

Full write-up: https://t.co/P4X7z7DTUU

[Justin Crites @dreambase_jc]

@supabase @DreambaseAI It's also changing how we build. We flipped the pipeline: the agent explores the data first with read-only SQL, then snapshots only what its answer depends on.

We're calling these just-in-time datasets. Still a prototype, but it did even a little better in our benchmark.

[🤓 Andy Keil @alwaysunday]

Holy sh*t, the new Supabase Compute is fast 💨 and you're gonna love it.

We ran some quick evals with early access (thanks @mattjohnstondev & @laktek) and the results are really promising.

💰 6.7x cheaper to run at ~$.05 per question (Woz's favorite) instead of our current $.36. That's roughly $500 for 10,000 questions vs the current $3,600.

With Compute, we're seeing 2.3x faster answers before we've started to optimize. And a bonus of 17x fewer tokens used. We like that.

@dreambase_jc is putting together a deeper dive early next week, and we'll be rolling it into @dreambaseAI right away. Anyone else tried it out yet?

[Justin Crites @dreambase_jc]

@posthog The only number I trust right now is how many people hit our site from one of their agents. PostHog helps with that, and we're tracking that number against an experiment to see what's working.

We just need to start seeing a meaningful increase and relate it!

[Justin Crites @dreambase_jc]

I hate AI vomit every morning telling me a story about what's working at my company. So I made a single place to view experiments, backed by Dreambase datasets and metrics, with actions built in to trigger anything from a slack message to spawning an agent to do follow on work.

[Justin Crites @dreambase_jc]

@posthog @LeSussyCat

[Peter Steinberger 🦞 @steipete]

OpenClaw deleted around 400k LOC of its own tests without much change in code coverage. Modern models just love writing tests for every tiny change, even if they aren't useful. This skill helped. https://t.co/a1jSBpfpns

[Justin Crites @dreambase_jc]

@alwaysunday @DreambaseAI I’ll be there! Love that it’s next to ABP, can hit my old stomping grounds

[Justin Crites @dreambase_jc]

@nateberkopec @thorstenball I agree, I've seen smaller codebases easily reach above 10k unit tests with no real upside. It just makes the code feel more rigid in an age where code is cheap to write and delete.

Are we moving back to BDD maybe?

[saila @sailaunderscore]

When you rely on LLMS: braindead slop-cannon

When I rely on LLMs: on the bleeding edge of cognitive offloading

[Justin Crites @dreambase_jc]

@posthog Been working with the scouts a bit and enjoying it so far!

I'd love to see a thread of where you all see the edges at. What isn't working so well would be a great mirror to this thread so we know where the boundaries lie.

[🤓 Andy Keil @alwaysunday]

New shirts just dropped and I'm so freaking stoked on them 🔥

We've been playing with some new brand elements and I love where we landed. This is just a taste. You'll see a lot more of it at @supabase Select next week.

If you'll be there, let me know and I'll hold one for you!

Now who wants to guess what we put on the back? 👀

[Justin Crites @dreambase_jc]

@lennysan @clairevo @DreambaseAI I'm looking for other folks who want to try it out themselves. Sign up for Dreambase and register for early access, and shoot me a DM or connect and I can get you in sooner.

[Justin Crites @dreambase_jc]

Last week I attended the @lennysan conference. That morning @clairevo asked the audience: "how many big experiments are you running every month"

That stuck with me.

[Justin Crites @dreambase_jc]

@lennysan @clairevo @DreambaseAI Then it adds automations on top to track the experiment against the metrics you care about. Now I get updates on how my experiments are going without having to do the checks myself.