michael litman

Culture, tech, and taste. Brand and product strategy. Signals before they’re obvious. Writing, building, decoding. Sneakerhead. Rekt. Doodles. Meebits. #NFFC

[michael litman @mlitman]

tl;dr i wrote a book over the last year published by BCS about web3, culture, art, & digital ownership

you can get a physical copy on amazon and many others internationally, kindle or audiobook on audible

https://t.co/TPugpZdFM1

my son tells me it’s a very digestible read

[GordonGoner (Wylie Aronow) 🍌 @GordonGoner]

New NFT book that quoted some of my thoughts from this last year. Gonna check it out. Thanks Michael ❤️

[michael litman @mlitman]

@punk5406 @lokithebird Freeeeee

[Garry Tan @garrytan]

High agency high taste is the unlock these days

[michael litman @mlitman]

@katiewav this is so great! I did something similar over at https://t.co/vuFvYDBh7S and it was a lot of fun!

[michael litman @mlitman]

The bottleneck moved. Move with it. Or optimise the wrong thing.

Full deck (23 slides): https://t.co/NZ8pjAWVZ6

Blog (1,339 words): https://t.co/n6tD6ugASs

Or: https://t.co/8PsXLPiIT4

23/23

Cred to the @AnthropicAI team for the day. Thank you @bcherny @katelyn_lesse @angjiang @_catwu and everyone else involved.

[michael litman @mlitman]

Each platform release this year answers one of the five bottlenecks.

Claude Managed Agents: scoped agent identity (bottleneck 3)

Ravi Trivedi's "Memory and dreaming for self-learning agents": context-bearing agents (bottlenecks 1 and 2)

Ralph Ramos's "What's new in Claude Code" + Punit Shah's "Getting more out of the Claude Platform": a codebase the model can reason about (bottleneck 4)

Cred to @bcherny @_catwu @angjiang @katelyn_lesse.

17/23

[michael litman @mlitman]

What to do. Three moves.

1. Audit one process for the new bottleneck.

2. Name your new bottleneck. Pick one of the five. Make it a named workstream with a named owner.

3. Run one thing your stack couldn't do until recently.

22/23

[michael litman @mlitman]

From @protocol7 on the @SpotifyEng main stage:

"Coding is no longer the constraint. Scaling devex to teams and agents."

The session title was the read. The constraint moved.

16/23

[michael litman @mlitman]

My overview from Code with Claude London, Tuesday 19th May 2026.

The Bottleneck Moved.

The model is no longer the constraint. The organisation is.

23-slide deck: https://t.co/bqKGsPKZhf

1/23

[michael litman @mlitman]

The default isn't "I prompt Claude."

It's "Claude prompts Claude."

Routines, autofix, code review, dreaming, memory. Synchronous coding is now a slice.

15/23

[michael litman @mlitman]

Bottleneck 05: Scaffolding readiness. Continuous.

Sid Bidasaria's "Stop babysitting your agents" + Margot van Laar's "The prompting playbook" picked at the same scab from different angles.

The harnesses, loops, instructions you built around earlier models are still in your stack. As Claude gets smarter, that scaffolding holds it back. Generalised primitives beat bespoke wrappers.

I have 28 hooks across my own workflows; several already rewritten or retired as new models needed less of them.

14/23

[michael litman @mlitman]

The reverse.

The orgs still optimising the coding bottleneck are buying more seats for a tool that just got cheaper.

The constraint moved underneath them.

13/23

[michael litman @mlitman]

Code got cheap.

Judgement didn't.

12/23

[michael litman @mlitman]

Bottleneck 04: Standardisation. 2-3 year programme.

@SpotifyEng spent a decade pushing engineers onto a smaller set of approved technologies.

What started as a developer experience play turned out to be the strongest moat for AI agents in their codebase.

11/23

[michael litman @mlitman]

Bottleneck 03: Identity. 2-year problem.

Identity is the unsolved problem of agent-native systems. Agent calls agent calls service writes a record. Who's allowed to do what.

@mondaydotcom is rebuilding its permission model for agents-as-users. Anthropic answers from the other side with Claude Managed Agents.

10/23

[michael litman @mlitman]

Bottleneck 02: Decisions. This year.

Alex Kaluzny (@doctolib CTO): prototyping features in actual production codebase. Weeks to minutes.

Result is not faster product. Too many viable ideas, leadership can't tell them apart fast enough. @ambricken "The thinking lever" research talk pointed at the cognitive side.

Taste becomes the floor.

9/23

[michael litman @mlitman]

Bottleneck 01: Review. This quarter.

Old model: a human reads every diff.
New model: humans pick what to read.

With agents writing most lines, review queue depth becomes the floor on shipping.

@SpotifyEng and @AnthropicAI are already running this model: auto-merge low-risk diffs, humans on the judgement calls.

8/23

[michael litman @mlitman]

Five new bottlenecks.

Ranked by what bites first.

01 Review: this quarter
02 Decisions: this year
03 Identity: 2-year problem
04 Standardisation: 2-3 year programme
05 Scaffolding: continuous

7/23

[michael litman @mlitman]

+76%.

That's Spotify's PR frequency after Opus 4.5 (h/t @SpotifyEng).

The bottleneck isn't holding. It's gone.

6/23