Twitxy Saved

Gleb Sidora

@GlebSidora

Co-founder & CTO @codos_ai
Backed by a16z @speedrun (SR007)

ex-Meta
San Francisco instrumentalzen.substack.com Joined March 2026
47 Following 155 Followers 64 Posts

Latest posts 5

Gleb Sidora @GlebSidora
Courage & vision is what usually stands between company-as-it-now and company-as-it-should-be
Dima Verified account @dimakhanarin Oct 30
95% of AI pilots never reach the P&L.

The companies that do have 4 pillars:
1. Goals that can't be hit without AI
2. Leaders who lead by example
3. 100% of company knowledge captured so agents can use it
4. Willing to disrupt themselves

Prerequiste: an AGI-pilled CEO who understands that the endgame of the AI-transformation is RSI ⬇️

Imagine Fable-6 or GPT-7 comes out. Would your company get better?
View quoted post on X
Gleb Sidora @GlebSidora
Another case where my deep-seated instincts backed by 15 years of industry experience turned out to be wrong
Kun Chen Verified account @kunchenguid Oct 8
ok it's official... AI-written unit tests and integration tests are empirically proven to be unhelpful. you should tell your agents to stop writing tests by themselves

on the deepswe eval set, banning sonnet 5.5 high from writing any tests actually resulted in slightly higher success rate (non stat-sig), with less time and token spent (stat-sig). this is pretty hard empirical evidence

across the tests that were written by the "tests allowed" baseline arm, 65% of them were unit tests, 35% were integration tests, neither bucket resulted in any improvement compared to not writing any tests at all

i also picked a random sample of 44 tasks subset where i completely disabled executing even existing tests - it also did not affect success rate at all

i spot checked many tests written in the baseline arm, and my intuition is that most tests are simply a repetition of the implementation

agent-written tests do not add any value because both the implementation and the tests were simply the agent's interpretation of our intent. the tests aren't any more accurate than the implementation itself

i suspect we can still extract some value from unit tests and integration tests if we describe them ourselves when we believe we can articulate our intent better through test cases than through requirements. but i have not proven this yet

also worth noting, during deepswe eval the agent wrote almost no e2e tests (only 17, compared to 3000+ unit/integration tests written). so this analysis does not prove nor disprove the value of e2e tests - i will do another eval specifically for that
View quoted post on X
Gleb Sidora @GlebSidora
Proud to be a part of @codosai!
Dima Verified account @dimakhanarin Oct 22
Most company brains fall apart at enterprise scale.

We just took #1 on the EnterpriseRAG-Bench: 92.64% at 500k files.

1.5 years ago I dreamt of us solving AI's memory problem.

Today our brain learns, remembers, and powers the AI transformation for every client.

But a brain is nothing without data. The companies winning are going after 100% token coverage, meaning every doc, thread, and system is visible to the AI.

How much of your company's knowledge can your AI actually see today? 👇

Kudos to @GlebSidora @OjasMor @RHLSTHRM 🎉

codos.ai/blog/aethos-en…
View quoted post on X