Timur

Two decades building for clients. Now building my own. Sharing the real process — the AI tools, the experiments, and the hard-won lessons.

[Timur @timurbuild]

@rubenhassid The time is up, the rug has been pulled…

[Timur @timurbuild]

@filippkowalski The leason is: invest in local LLMs, that’s the only way. Open-source will save the world once again

[Timur @timurbuild]

Anthropic’s takedown of Fable 5 by US government order is one of the best illustrations of why local LLMs are the only way to go.

[Timur @timurbuild]

@filippkowalski That’s how the future might look like: they will persuade masses stop learning to code and then they’ll force people to pay for every token, there will be no subscriptions.

[Timur @timurbuild]

@maks6361 Cyprus?

[Timur @timurbuild]

I like this new Opus 4.8 honesty.

[Timur @timurbuild]

@BHolmesDev @anothercohen Just saw your tweet while looking for a solution. Thanks for the guidance!

[Timur @timurbuild]

@mikehostetler When I hear the word "Agent," I can only think of one:

[Timur @timurbuild]

@_aurumX It looks like they're flooded with AI slop apps.

[Timur @timurbuild]

I asked @claudeai to give me his take on your article. Here's what I got:

`Honest take: the article mixes one real signal with a lot of marketing.

What's real:
- Demis Hassabis genuinely says ~2030. He's been consistent, and he's not a hype merchant — DeepMind tends to under-promise relative to OpenAI.
- AGI capability gains since 2022 are unprecedented in the field's history. Timelines have compressed.
- The career advice — taste, judgment, relationships — is sound regardless of timeline.

What's overhyped:
- "AGI" has no agreed definition. Demis means something specific (matches human cognitive breadth). Sam Altman means "useful at most economic tasks." The press collapses both into "smarter than humans." Different bets.
- Sample of one. Yann LeCun (Meta) says decades and current LLMs are a dead-end road to AGI. François Chollet (ARC prize) says we're missing core capabilities. Gary Marcus says never on this paradigm. Picking Demis's number alone is cherry-picking.
- "Maximum 4 years" is a confidence claim no one can back. AI forecasting has a brutal track record both directions (self-driving was 2 years away in 2016).
- The frame — "shook his hand, here's what it means for your job" — is a rhetorical move. A 2-minute hallway chat at I/O isn't insider access; it's PR.
- The author's business model rewards urgency. Newsletter → Substack → consulting funnel works better when readers feel the clock ticking.

The useful reframe: The advice ("be the person worth having lunch with, develop taste, take responsibility") is goodvwhether AGI arrives in 2030 or 2050 or never. That's the giveaway — it's a hedge dressed as a prediction. If thevprescription survives the prediction being wrong, the prediction was decoration.

My honest probability: meaningful AI capability disruption to knowledge work by 2030 — high. Demis's specific AGIvdefinition by 2030 — maybe 25-35%. Civilization-rewriting Singularity by 2030 — low. The article treats all three as thevsame event.

Read it for the career advice. Discount the timeline by ~50%.`

@rubenhassid what do you think?

[Timur @timurbuild]

@simonw There’s no excuse to not to have tests nowadays…

[Timur @timurbuild]

@zodchiii Why?

[Timur @timurbuild]

@josevalim I’m afraid that thinking process is dead. Where are all the people and their thoughts? Posts and articles written by humans? It’s starting to feel lonely here…

[Timur @timurbuild]

@GoogleDeepMind So you're going to share your screen with Gemini? Say goodbye to your privacy...

[Timur @timurbuild]

My morning reflection: We are entering an era in which thinking is becoming optional, robbing us of the capacity to form thoughts of our own. Beware!

[Timur @timurbuild]

@hooeem Enjoyed reading the article, but honestly, what do you get in the end? A pile of narrow-scoped documents that are going to drive your decisions? I don't know; it feels to me like productive procrastination rather than doing something that will increase one's output.

[Timur @timurbuild]

Hey @conductor_build, is there any way to annotate files and send feedback back to the agent? That’s the only thing I’m missing in Conductor.

[Timur @timurbuild]

@mikehostetler @jacob_luetzow Keep up the great work. Love your laid-back style!

[Timur @timurbuild]

@tibo_maker Easier said than done, unfortunately… But ofc makes total sense.

[Timur @timurbuild]

@_joeygold Congrats! What are you building, is it an ios app?