Adnit Kadrii

[Adnit Kadrii @adnitThe1st]

@rezoundous

[Adnit Kadrii @adnitThe1st]

@mikeydsoftware Sol 6 is so obstinate that it will keep doing its own thing, even if you try to explain it 10 times.

[Adnit Kadrii @adnitThe1st]

@kylegawley Are you accepting drivers? I can drop everything on the condition i get 70% a ride.

[Adnit Kadrii @adnitThe1st]

@Ananth7e Pro Max Ultra + when?

[Adnit Kadrii @adnitThe1st]

@designwkarthick Let me push you over the edge.

[Adnit Kadrii @adnitThe1st]

@alicanvergin @thsottiaux I have indeed gotten addicted to resets. @thsottiaux still haven’t received my fix.
Withdrawal symptoms already kicking in.

[Adnit Kadrii @adnitThe1st]

Lately, GPT-6 Astra seems to over complicate even simple commands.

[Adnit Kadrii @adnitThe1st]

@sama Disappointed, @thsottiaux @sama @jxnlco. My 7-post submission documented a real multi-agent workflow: specification, adversarial review, revision, implementation and independent code review. The dataset has 9,984 records, 16 below the advertised 10,000. I’d appreciate a review.

[Adnit Kadrii @adnitThe1st]

@Jonathan1294931 @sama Same

[Adnit Kadrii @adnitThe1st]

@egorkabantsov Bad timing my friend. At least you had 10 hours with fable 5

[Adnit Kadrii @adnitThe1st]

Removing it before the communicated deadline, with no proper migration window, is unacceptable.
2/2 @claudeai

[Adnit Kadrii @adnitThe1st]

Anthropic told subscribers that Claude Fable 5 would remain available until July 19, 2026.

It was removed today, July 17, at around 20:47 CEST—two days earlier than announced.

Users paying $100–$200 per month built serious workflows around this model.
1/2 @AnthropicAI

[Adnit Kadrii @adnitThe1st]

The biggest productivity gain hasn’t come from choosing one “best” model.

It comes from assigning specialized agents different roles, making their reviews independent, and forcing them to challenge one another before I trust the result.
7/7

[Adnit Kadrii @adnitThe1st]

What I love most about GPT-5.6 Sol isn’t just its implementation quality, it’s how effective it is as an adversarial reviewer.

My current workflow uses Claude Fable 5 as the orchestrator and Sol through Codex CLI as the critic and implementer. 🧵
1/7

[Adnit Kadrii @adnitThe1st]

After implementation, Fable reviews the code itself and simultaneously spawns another fresh Sol CLI instance to review the implementation independently from scratch.

If either finds an issue, it gets fixed and reviewed again.
6/7

[Adnit Kadrii @adnitThe1st]

After implementation, Fable reviews the code itself and simultaneously spawns another fresh Sol CLI instance to review the implementation independently from scratch.

If either finds an issue, it gets fixed and reviewed again.

[Adnit Kadrii @adnitThe1st]

Only then does Fable spawn a new GPT-5.6 Sol Codex CLI instance to implement the feature.

Fable remains the orchestrator. Sol acts as the independent critic first, then as the implementation agent.