Jessica Kerr

Symmathecist, @honeycombio. she/her
Tweets are mine, license CC0. It is not about me.

[Jessica Kerr @jessitron]

I don’t wanna “build software” so much as

build understanding and express it in software.

[Jessica Kerr @jessitron]

"achievement" unlocked: send an LLM the complete works of Shakespeare

[Jessica Kerr @jessitron]

Verum ipsum factum: we can only truly know what we ourselves created. (Vico)

This rings true in software and it scares us rn

[Jessica Kerr @jessitron]

With technological acceleration, "actual locations tend to become 'non-lieux', without history, identity, or relation" - Rosa

[Jessica Kerr @jessitron]

Short story idea:
You can go into the bathroom at Starbucks and come out in any other Starbucks.
But you can never leave Starbucks.

[RubyConf @rubyconf]

Final keynote announcement for RubyConf 2026: Jessica Kerr (@jessitron) closes Day 1 on July 14 at 5:00PM.

25 years in software. 14 years keynoting conferences around the world. One community she calls home.

We are so glad to have her on stage.
https://t.co/LOA2ZHVikc

[Jessica Kerr @jessitron]

I have found the killer app for Granola.
It takes notes during the meeting. Afterward, I ask it

"did I sound like an idiot in this meeting?"

of course it says no, in detail, and I feel reassured

[Jessica Kerr @jessitron]

Our future is sorta based on the last 30 years of internet content.

LLMs are possible because of the corpus of data that is the Internet of 1995-2023. More recent data is sus bc LLM content.

AI is super useful and a big part of our future.

Makes me wish I had written more.

[Jessica Kerr @jessitron]

"most people think they are a better-than-average driver"

They could be right!
with some terrrrible drivers skewing the average below the median.

[Jessica Kerr @jessitron]

OH: "corporate decided to switch to online training, and now every new person sucks at their job."

Work as planned never matches work as done

[Jessica Kerr @jessitron]

LLM conversation context is like Katamari: it gets bigger and bigger, picking up everything it finds as it goes

[Jessica Kerr @jessitron]

Metrics are an agentic dead end.
@QuinnyPig #o11ycon

[Jessica Kerr @jessitron]

Add a tag to your metrics:

“direction”: “higher_is_better”
- @StephWangBuilds from @MongoDB #o11ycon

ah, that looks like a great idea!
Better for agents, better for humans

[Jessica Kerr @jessitron]

Evals are one of the hard parts. How do you define “good” and how do you test it — do you freeze conditions to test just the model?
but the tools and data available have changed, and those are parts of the agent too.
Eval scenarios have to change as the system does

[Jessica Kerr @jessitron]

Stripe’s observability team wrote an agent to help respond to incidents.

“Some early quick wins made us think it would be easy. It wasn’t”

- Rob Miles, Mike Cowgill #o11ycon

[Jessica Kerr @jessitron]

@jessephelps yeah! that sounds fun

[Jessica Kerr @jessitron]

LLMs give us a new point of contrast for "what makes us human?"
LLMs can do reasoning. That isn't what makes us human.
LLMs aren't conscious, though. They don't have experiences, and they don't connect with each other.

[Jessica Kerr @jessitron]

Pull requests are as relevant as ever: should we make this change in our software?
Code isn’t the central artifact anymore.

- @depoll #o11ycon

[Jessica Kerr @jessitron]

Coding agents and how they work, from the ground up, just what you need to know to use them effectively.
That's the course that @avdi and I are making at @GracefulDotDev starting now.

https://t.co/vsnXcJFhzP