Michael Bleigh

Building the servers for serverless @Firebase. Formerly founder @divshot. Web platform maximalist.

[Michael Bleigh @mbleigh]

@josevalim Hooks should have been built as part of MCP from the start.

[Michael Bleigh @mbleigh]

One pattern I find myself reaching for frequently these days is to gather data into a heavy JSON "kitchen sink included" format and then build lightweight CLI renderers that slice and dice the data into various token-efficient text views for an agent to understand.

[Michael Bleigh @mbleigh]

@Noahpinion The only true signal that a franchise should stop is multiple failed commercial outcomes, which usually arise 3+ attempts after the failed artistic outcomes begin.

What I've realized is that I don't actually have to watch things I don't think I'm going to enjoy! Problem solved.

[Michael Bleigh @mbleigh]

This is a blessing and a curse. I sometimes find myself starting from scratch over and over, never committing to "good enough" bc I know how easy it is to do the next version.

Sunk cost is a fallacy but I think sometimes it's also a useful heuristic to force you into finishing.

[Angie Jones @techgirl1908]

what I love about working with coding agents is that starting over is essentially free.

after implementation, I can objectively review and if I'm unsatisfied, I can just revert everything and start afresh... with new lessons learned.

if I implemented the solution by hand and invested hours into going down the wrong path, it would be harder to convince myself to just throw it away, ya know

[Michael Bleigh @mbleigh]

@willmcgugan @AnthropicAI Polish can be a differentiator in mature markets, and AI is pretty great at fixing these kinds of small polish issues. But coding agents aren't a mature market - they're probably the most rapidly evolving product category in history.

[Michael Bleigh @mbleigh]

@GergelyOrosz If you told someone in 2023 that a $300M price tag was an acquihire they'd laugh in your face.

[Michael Bleigh @mbleigh]

This actually works as a bidirectional grading system for evaluating agents as well. If Agent X can get it done smoothly from the prompt but Agent Y can't, the service or the agent or both have work to do.

[Michael Bleigh @mbleigh]

My new bar for onboarding is "give me a prompt I can paste into a halfway decent agent that gets me completely set up and onboarded to your service with no prior knowledge of how anything works".

Results are graded on a scale of how much work I have to do instead of my agent.

[Michael Bleigh @mbleigh]

3 things every LLM API ought to have knowing what we know now about building agents:

1. Allow add'l function defs without busting prefix cache
2. Interleaved "developer" messages with stronger instruct weight
3. Canonize XML-tag-ish structure into first-class "section" primitive

[Michael Bleigh @mbleigh]

Waiting for a reply shaped like this on this post:

"A slop button changes the game on automated posters. One click and the post is gone. It's not just cleaning up the timeline, it's changing the incentive structure."

🙃

[Michael Bleigh @mbleigh]

Every social media service needs:

1. A "slop" button that is a combination user mute and report of a post as being low-quality.
2. ToS allowing discretionary banning of users for automated low-quality posts.

Right now slop replies are clearly rewarded. They need to be punished.

[Michael Bleigh @mbleigh]

@Noahpinion Slop means "low-effort, mimicking the fat part of the distribution".

Last Jedi is neither of those things. People hate it because it tried an out-of-distribution take on a mainstream property and didn't really pull it off.

[Michael Bleigh @mbleigh]

@MattBruenig @TheStalwart @DKThomp @jdcmedlock I could dedicate my life to being the best in the world at bashing my head against the wall, work 18 hours a day for 10 years. Best head basher on the planet.

Who else benefits from my effort?

George Lucas made a thing that millions of people love, seems worth rewarding.

[Michael Bleigh @mbleigh]

The dark part is imagining what we'd hook up to "pain" receptors in conscious machines.

Lint failure in its code gives it a little jolt, security hole puts it in the "hospital".

But even things like pain and time perception are likely to be "simulated" soon. Where's the line?

I'd posit that we will never be able to establish a bright line test for consciousness and recognition of artificial personhood eventually becomes a divisive civil rights issue.

[Michael Bleigh @mbleigh]

@samuelcolvin Check out https://t.co/7RUUhVmsNB

[Michael Bleigh @mbleigh]

@StatisticsFTW I think it's the biggest failure of GraphQL docs and ecosystem onboarding that this isn't in the first 10 minutes of being introduced to GraphQL.

It's like teaching people how to use guns without telling them safeties exist.

[Michael Bleigh @mbleigh]

Yes AI will be able to do things with "competent" taste because that's the fat part of the distribution.

But to do something that *innovates* on design and experience you need an insight that is inherently out of distribution. That's the potential moat.

Regardless, humans with discerning taste will be able to get a lot more out of AI even as AI becomes much more capable.

[Michael Bleigh @mbleigh]

Okay so it turns out Telegram is 1000x easier to set up with a bot than WhatsApp or Google Chat or SMS.

I don't use Telegram generally, I didn't particularly want to start, but I might for this reason alone.

[Michael Bleigh @mbleigh]

I wanted to play around with building a personal agent for my family and the process of getting working credentials to talk to a chat app is...next to impossible?

Is there some secret shortcut to doing this that doesn't require me to register an LLC or something?

[Michael Bleigh @mbleigh]

@rockwotj @TaskletAI But I want to hack it together myself out of bubblegum and duct tape!