Are you building or buying Evals?
11.02.2026 01:08
👍 3
🔁 0
💬 0
📌 0
Are you building or buying Evals?
I take this back.
A well crafted MCP can mask a poorly defined API pretty effectively.
Don't ask me how I know...
The growth of AI Assistants and usage is forcing a tighter focus on APIs as a product.
Evolution over time and a lack of "incentive" to consider even core APIs as product really becomes obvious (and problematic) when an AI Assistants needs to understand and use them effectively.
A fun exploration of cache effectiveness for reverse-geocoded data in Rover Search.
www.rover.com/blog/enginee...
How do we feel about starless?
#astrophotography
The Orion Nebula, from my sessions last week.
dylanbernard.com/astrophotogr...
Finally got some clear nights!
dylanbernard.com/astrophotogr...
Great writeup about how we mitigate risk for core Search infrastructure upgrades at Rover.
From 5.6 to 8.x: Upgrading Elasticsearch at Scale
www.rover.com/blog/enginee...
Finally, some clear nights.
(I promise "night sight" makes the light pollution look way worse than it is)
Curious what people have run into when adapting existing core services to support AI use cases (agents, MCP-style tool interfaces, LLM-driven interactions, etc.).
What surprised you? What broke in weird ways? What do you wish you’d designed earlier?
Someone is in the holiday spirit.
Totally! And yes, definitely give Claude a shot if you can. Agent Skills have been super impactful for me so far. I'm tinkering with a Skill now that updates itself with summary learned business logic as it goes to cut back on the need to re-learn on future invocations. Effectiveness of that TBD.
2025 has been the first year where I REALLY embedded AI tooling into my workflows. Here's a quick brain dump around how AI tooling is (and is NOT) working for me right now.
Let me know about your experiences as well!
dylanbernard.com/ai/2025/12/1...
This was a tricky one! Diagnosing (seemingly) random, low volume 10x performance regressions on Rover Search.
www.rover.com/blog/enginee...
I re-watched Krull last week. The 80s gave us some truly iconic movies. Think of all the weird movies we could have today if the DVD market and movie theaters didn't collapse under the weight of streaming.
Claude Agent Skills are super cool. Having a lot of success organizing more intricate workflows this way!
I finally pulled the trigger on a Niche Zero grinder as part of my "endgame" espresso set up and holy cow this thing is awesome.
Are we still doing starter packs?
Put this one together because I love seeing things that lovely folks write on the internet, and I'm sure there are more people to meet and add to this list.
go.bsky.app/AnM2t7r
Check out how we got started with AI coding agents at Rover!
www.rover.com/blog/enginee...
I know it's at least not uncommon for coding bootcamps to strongly encourage (or even require) students to reach out to practicing devs as a networking exercise. Maybe that's propagated beyond bootcamps now?
I field a few of these regularly, though not at the volume above.
Basically everyone is running some sort of AI adoption / advocacy programs within their tech organizations.
Is anyone also running some sort of "AI Responsibility" or "ethics" programs? How's that going? What's working for you?
I'm always looking for ways in which my "Personal AI Principles" are pressure-tested by reality.
Recently, for example, I've found that I'm willing to bend on my, "Plan and edit, but do not write using AI" principle specifically for technical documentation.
theproductstaffeng.com/ai,/personal...
Finally committing to scheduling my sabbatical. Going to spend 8 glorious weeks gardening and pointing my camera at the stars next spring.
First 24 hours away from the kid this weekend.
Excited? Yes.
Will I be scrolling through baby pictures on my phone after a cocktail tonight? Also yes.
As I use AI tools more in both my personal and professional lives, I've felt the need to commit my *personal* AI principles to writing.
I've shared with a few coworkers and have been surprised by the positive reactions.
Let me know if you have thoughts!
theproductstaffeng.com/ai,/personal...
So it turns out having a kid really takes over your free time.
So, 8 months later, here is part two of my Technical Experimentation series focused on building and maintaining trust.
Trust is a core tenet of healthy technical experimentation. It is hard earned and easily lost. Read more below!
The Twilight Zone is one of those things I've always known I would probably enjoy, but never actually got around to.
So anyways, I'm like 4 episodes in and (shocking, I know) turns out it's really good.
The Vast of Night is such a solid little flick.
Turns out that choosing astronomy-based hobbies in a geography where the sun sets at like 9pm all summer, and the other 3/4 of the year is calling for rain is not the best idea.
Anyways, let's buy that new telescope I've had on my wishlist for a year...
It's really easy to "forget" to allocate time to cleaning up feature flags after wrapping up product work.
Three years and 800 stale flags rotting in your codebase later, you'll really appreciate the hour or two that you saved (/s).