Dmytro Budym

SDET Lead @ Contentsquare · AI-powered test automation
Playwright & TypeScript · Agentic QA workflows 🤖
Insights →

[Dmytro Budym @dbudym]

https://t.co/BMhg4gmOPp

[Dmytro Budym @dbudym]

Just saw in my terminal📟
"Meet Fable 5, our newest
model for complex, long-running work."

Stripe migrated 50M lines of code in 1 day. Manually: 2 months ⚡

Let's burn some @AnthropicAI tokens😀

[Dmytro Budym @dbudym]

@bcherny Great feature 🔥

[Dmytro Budym @dbudym]

Are you still prompting turn by turn, or have you started setting goals and walking away?
And if you've run /goal in prod — what broke first, your condition or your trust? #AIEngineering

[Dmytro Budym @dbudym]

I stopped prompting Claude Code.
I gave it a goal instead.
"Raise test coverage to 80%. Don't stop until it's done."
Then I went and had lunch.
That was not a metaphor.

[Dmytro Budym @dbudym]

Early adopters are calling /goal "the most underrated AI feature of 2026."
I think that's right. Not because it's flashy. Because the mental model shift is quiet and then suddenly enormous.
You don't feel it until the session closes and the work is just... there.

[Dmytro Budym @dbudym]

I've been in test automation for years. The shift from "write this test" to "own this coverage target" is not incremental.
It's the first time I've genuinely felt like I handed off a QA task rather than assisted with one.
That's a different relationship with AI. →

[Dmytro Budym @dbudym]

This is a different mode of working entirely.
Prompt-by-prompt: you stay in the loop, you guide, you correct.
/goal: you define the outcome, you disappear, you come back to a ◎ achieved state.
One is using a tool. The other is delegating to a worker. →

[Dmytro Budym @dbudym]

One practical caveat they don't put in the headline:
Running /goal with agents active burns quota roughly 10x faster than a normal session.
Set a turn limit inside the condition itself "or stop after 30 turns" if you want a hard budget cap. The evaluator tracks it. →

[Dmytro Budym @dbudym]

The failure mode isn't the model.
It's the condition you wrote.
Vague goal = evaluator never confident = session runs forever.
"Fix the auth bug" will loop indefinitely. "All tests in test/auth pass and lint is clean" will close cleanly. →

[Dmytro Budym @dbudym]

But here's the part nobody talks about:
The evaluator can only read the conversation transcript. It doesn't open files independently or run commands on its own.
So your Claude can prove inside the session. "npm test exits 0" works. "Everything looks good" doesn't. →

[Dmytro Budym @dbudym]

What it looks like in practice:
You run /goal. A "◎ /goal active" indicator appears.
Check status at any point: elapsed time, turn count, token spend, and the evaluator's last reason for not closing.
I checked mine after 40 minutes. Turn 11. Still running. →

[Dmytro Budym @dbudym]

That separation matters more than it sounds. When a single model both does the work and decides when to stop, it drifts. It convinces itself it's finished.
The evaluator is a cold second opinion after every single turn. →

[Dmytro Budym @dbudym]

The architecture is the interesting part.
After every turn, a small fast Haiku model acts as an independent evaluator. One question: has the condition been met?
No? Claude starts another turn.
Yes? Goal marked achieved.
The working model never decides when it's done. →

[Dmytro Budym @dbudym]

Claude Code shipped /goal command two weeks ago.
You type a completion condition - a real, verifiable end state and Claude keeps working turn after turn until a separate model confirms the condition is met.
Not "do your best." Done when it's actually done. →

[Dmytro Budym @dbudym]

Stop stuffing your AI agents with context. 🛑

Adding auto-generated instructions can hike your inference costs by 20% while actually decreasing success rates. 📉

More context ≠ more intelligence.

https://t.co/sCAppfRxUA

[Joe Colantonio @joecolantonio]

🚀 #AG2024 kicks off in a few hours! Join us for game-changing insights into #testautomation, AI, DevOps, and more. 👉https://t.co/cFrWiUCn48 @lazyautomator @Applitools @mablhq @jonathon_wright @Keysight @MushHonda @KatalonPlatform @ArtemBondar @d_budim @AmSachinGaikwad

[Joe Colantonio @joecolantonio]

@d_budim brings his expertise to #AutomationGuild2024 with #CleanCode in #TestAutomation. Discover the keys to building reliable, maintainable test suites that speed up #softwaredelivery and boost team collaboration. Join us! 👉 https://t.co/cFrWiUCn48 #AG2024

[Dmytro Budym @dbudym]

Honored to be a part of Automation Guild 2024, thanks to @joecolantonio incredible effort. This event promises to be a landmark in test automation. Eagerly awaiting to connect and contribute!

[TestGuild @testguilds]

@d_budim's session at #AutomationGuild2024 is a must-see! Get ready for groundbreaking ideas on how to have clean code in #testautomation. Reserve your spot! 👉 https://t.co/1w5uq2iqAR #AG2024

[Dmytro Budym @dbudym]

@eliasnogueira @Java_Champions Congrats and nice stuff 😀