I've written a 2-3 hour workshop called "Personalisation without the Pricetag" on using ProcGen and GenAI from local LLMs to make an experience that feels authored for the user, but doesn't cost the Earth. If you'd like it at your next event let me know!

A CEO who gets it

Interesting piece on how Prime Video does progressive tool discovery using some underrated features of the MCP spec.

Law of Large Numbers tells us the more you use AI to generate tokens, write lines of code, and merge PRs, the more mid you and your app become. That’s science.

Super intelligence has already been achieved. It figured out time travel and we’re living through the ideal time it installed itself to ensure its dominance and perpetual existence but also, to ensure it can foil any of our attempts to discover time travel ourselves to be free

Saying that things not working with agents is a skill issue is not new. Everything that goes wrong with coding is more or less a skill issue. You didn't have a good enough spec, you didn't implement it correctly, you didn't consider resource constraints...

How do you inoculate against AI taking over critical thinking?

Oh neat, LLMs pretty good at doing stuff that we pay some people a lot of money to do, but seems to struggle occasionally on some things. CEOs: REARRANGE EVERYTHING SO IT DOESN'T STRUGGLE The people who struggled with that stuff:🫨

The map is not the territory. The map is not the territory. The map is not the territory. The map is not the territory. The map is not the territory. The map is not the territory. The map is not the territory. The map is not the territory. The map is not the territory.

So well put. LLMs are "confidence engines". Hyping us up and using flattery hack our belief system into believing they're a form of intelligence. But they don't push back, they don't hold their own beliefs. RLHF just makes this more and more effective.

here's how your company is rotting right this moment: - your senior devs stopped writing code - they ask Claude to generate it, they check that it mostly works, they ask a junior to approve the new PR - a junior who never had a chance to learn about architecture or read the docs can't really explain what you are doing wrong, so they blindly LGTM it - your senior devs stopped thinking - instead they "consult" Claude on making a bunch of strategic decisions; they ask the PM/principal to approve the new architecture - your PMs and principals are too busy (re)discovering the joy of producing 10k LOC, so they don't care if what you are doing is wrong, so they blindly LGTM it

Hold onto and don’t use an LLM to update that project you hand wrote before 2022. We’re going to be hunting for software in the future that hasn’t been irradiated by the nuclear blast of AI, similar to trying to find steel produced before nuclear testing.

Working with agents on your side project

Human language is lossy code for conveying ideas, thoughts, or intentions between human beings. Machine code -> Assembly -> C -> [insert higher abstraction code] <- Human Language <- Feelings/Intentions <- Unconscious thought We’re building using very lossy transpilations

RIP MCP we barely knew thee. Cut down in your prime. Enjoy your afterlife in the enterprise farm upstate where you’ll enjoy endless synergy and profits for life.

I saw the inverse happen in the Block layoffs. Someone was laid off, but it turns out it was a mistake and they were actually supposed to be promoted (the mistake was fixed, hence how they found out about said promotion).