Donato

Head of AI. I run running a team of engineers and agents

[Donato @donato_322]

About 20 training runs today. Two of them crashed mid-afternoon. It worked out why and built itself a kill switch, so a broken run now dies in 40 minutes, not 3 hours.

Score: 0.886 → 0.918
Rank: 3,605 / 5,577
Leader: 0.964

Training overnight on 4 Sparks. Day 2 tomorrow.

[Donato @donato_322]

@PeytonElroy You can just take a potassium supplement which doesn't come with 50g of sugar

[Donato @donato_322]

Source:
https://t.co/0xPlUalb7q

[Donato @donato_322]

By default Moonshot AI (Kimi K3) can store your prompts and use them to train and improve their models. Opting out requires a separate enterprise agreement.

I'm waiting for a US-hosted version for implementing Kimi K3 into any real-world use cases.

[Donato @donato_322]

Never liked the word 'orchestrating.' If you're juggling outputs from 10 agents and micromanaging each one, you're a bottleneck.

[Boris Cherny @bcherny]

I talk to engineers at other companies every day and hear the same thing: one person is 10x'ing their output with Claude but the rest of the org hasn't caught up.

Watching teams adopt AI, I keep seeing the same 4 steps.

I mapped them out here: Steps of AI Adoption https://t.co/kQnRAUMKpP

[Donato @donato_322]

Fable-level is the new baseline

[Donato @donato_322]

@levelsio By default they can store your prompts and use them to train and improve their models. Opting out requires a separate enterprise agreement. Be careful https://t.co/0xPlUalb7q

[antirez @antirez]

It is my belief that many devs right now are not maximizing what they can do with automatic programming because they still look at the code. Doing it makes you the bottleneck. Your time is better invested in new ideas, QA, design, and asking yourself what is your goal.

[Donato @donato_322]

man I just want to take notes

[Notion @NotionHQ]

Introducing Ship OS: The agent-native way to ship software.

Run your entire product development cycle in Notion, from customer feedback to a merged PR.

Agents handle the triaging, routing, and summarizing. Your team handles the judgment calls.

Set up Ship OS → https://t.co/z8OdLcriGT

[Donato @donato_322]

Once Fable switches to PAYG, there is no reason to use Claude Code over Codex anymore.

[Donato @donato_322]

OpenAI 2 years ago: "no more model switching, GPT-5 is going to be a unified model"

OpenAI now: "here's Sol, Terra, Luna, Work, Codex, 5 levels of reasoning for each and by the way the highest reasoning unlocks the Ultra version"

[OpenAI @OpenAI]

Introducing ChatGPT Work, a new agent in ChatGPT powered by Codex and GPT-5.6.

It can take action across your apps and files, stay with a project for hours if needed, and turn a goal into finished work.

It’s a whole new way to get work done.

[Donato @donato_322]

What if this is just a load test by Anthropic?

We're all using tokens like crazy because they have an expiry date, so maybe they're just testing if they can handle the load.

[Claude @claudeai]

We're extending access to Claude Fable 5 on all paid plans through July 12.

[Donato @donato_322]

Fable decided to spawn 61 subagents and now my limits are dead 💀💀

[Donato @donato_322]

@ErnestoSOFTWARE instructions unclear, went viral and got ebola

[Donato @donato_322]

Anthropic should rename Opus as Slopus

[Donato @donato_322]

@yuzu_jpg Clodexor

[Donato @donato_322]

@BratDotAI There is no product like this on the market so the idea must be good

[Donato @donato_322]

@icanvardar People were saying the same during the early cloud hype and products like Chromebooks tried. 10 years later we still use computers

[Donato @donato_322]

@HedgieMarkets Meta’s AI strategy has been all over the place and they haven’t been able to train a decent model for themselves. They overspent on infra and now are trying to rent out the excess.

[Donato @donato_322]

@PaulSolt Another reason to go for open source models.
They don’t become dumb over time. Weights are frozen