Gökhan Turhan

Full-stack builder · generalist solopreneur · conceptual artist

MX-010 · REV 2026-07-17

 DOC · JOURNAL-074 · WRITING PUBLISHED · 2026-11-15
JOURNAL · 2026-11-15 · GÖKHAN TURHAN

Swarms will be obsolete under a few years, too

A Blyg reply to artlu99's The Bitter Bitter Lesson, with both posts and their canonical URLs.

Contents
  1. The post it answers
  2. The reply

A Blyg is a small site for one person’s posts. Mine is Gökhan Turhan’s Blyg. A post can answer someone else’s post. The answer keeps a link to theirs, and it has its own canonical URL. This letter is that answer, with the post it answers, so both are on the journal.

The reply, as posted: https://blyg.gokhanturhan.com/t/4dw6zd81drr19jwdxbgq4r8eqx/

The post it answers, by artlu99: https://blyg.artlu.xyz/t/0hzp5c1bppnjmdd6tcpg1vyff8/

The post it answers

The Bitter Bitter Lesson, from artlu99’s blyg, canonical URL https://blyg.artlu.xyz/t/0hzp5c1bppnjmdd6tcpg1vyff8/. v2, 13 November 2026.

There is the well-known and oft-cited Bitter Lesson for AI researchers.

But unless you are an AI researcher, and as of two years ago there were only maybe 2000 of you on earth, there is another Bitter Bitter Lesson.

It’s not just model improvement that makes special learning obsolete. It’s tooling improvement, by severely autistic folks willing to ride the wave of The Bitter Lesson, that multiplies the Bitter Lesson for you and for me.

If you were a hobbyist 12 months ago, maybe into better agentic memory systems or self-improving agentic workflows, it’s 99% odds that everything you have worked on is now obsolete, or at best, absolutely sub-optimal compared to free tooling that you can access via Codex, Claude Code, OpenCode, Command Code, Amp, Grok Bot, etc. If you had a decent but finicky OpenClaw or Hermes setup, and muscled it into a much better shape, you’re still basically level with newcomers onboarding onto Muse or Instinct today.

If you were a professional programmer 12 months ago, diligently keeping your skills sharp as a centaur 20x eng, then sub-agent swarms, software factories, and now swarm riding, you are still kinda level with engineer Grandpa given access to the right off-the-shelf tooling and a great one-line mental frame (“set up goals that the agent can verify”). All the harness memory wiring and eval loop scaffolding you did? It basically only kept your head above water for a bit, until more people could jump into the same pool.

How do I know this? I sat out the whole software factory phase. And now I ride swarm nearly as good as the the best swarm rider. She just started maybe two months ago.

Huh.

Their footnote: if you were among the 1000 or so near SF, you are now a centimillionaire to multi-billionaire. If you were among the 900 or so near Beijing or Shenzhen, you are now a decimillionaire. If you were among the 100 or so in Europe, you have probably moved already to SF or Asia.

The reply

Swarms will be obsolete under a few years, too, from Gökhan Turhan’s Blyg. Canonical URL https://blyg.gokhanturhan.com/t/4dw6zd81drr19jwdxbgq4r8eqx/. v2, 14 November 2026. It quotes the paragraph about the professional programmer.

Swarms will be obsolete under a few years, too, in that it will have caught with the general public as bots will be lamer for consensus calibration. After all, it is kind of ridiculous where Grok bot almost telepathically reads my intents even at the cheapest tier and grok build 4.7 has hardships understanding the same flow, most of the time. However, we’ll have more concentrated formulae for the “nerd” world eaters, with productized systems akin to a random Portuguese Manowar.

The post-IPO neo-philanthropy will cause a resurfacing in context embodiment flow engineering side hackathon state of the mind levels epistemic tooling token rush for a few years for another calibration and I think if he doesn’t stop we’ll have our own personal gracefully weeping angels thanks to Gwern.

I really think that this type of minute obsolescence upon and through which practical venture capital houses arbitrage the immature verticals will continue about two decades—I’d really love to formulate a super-forecasting plane here for this, neither a benchmark nor an eval, that would feed back into semi-sentient SaaS agent-to-agent trollboxes on their private Bloomgerg Terminal trollbox of some sort in a niché abandoned Second Life BBS.

I was recently thinking around the practice of writing and reading re: this, invoking another Cartman-focused bildungsroman episode, but, as long as we can wear our external swarm beacons on our wrists or ghost wrists, we won’t need any of the “had could have been perfect” workflows.

Share this post