Mark Gerrard

AI orchestration in production: 9 model families, adversarial panels, nothing ships unverified. Writing on verification in a cheap-generation world.

[Mark Gerrard @MarkGerrardAI]

Anthropic's AI-native SDLC playbook names "an independent confidence gate between stages" and doesn't build it. ARB (Agent Redis Bridge) is that gate: harness-agnostic, model-agnostic, nine model families, author never gets a vote. Outline 👇 https://t.co/cP7Xuryu5f

[Mark Gerrard @MarkGerrardAI]

Has Sol been moved down to Sonnet level (replacing Terra) and Astra Minor (when released) to become the Opus level model?

[Mark Gerrard @MarkGerrardAI]

Congrats to you and the team on an awesome release!

Is there any guidance on when Fable 5.1 should be used / performs better than Opus 5.5?

TIA

[Boris Cherny @bcherny]

Opus 5.5 is a really good model. It's been my daily driver the last few weeks.

We had Opus 5.5 and Fable 5.1 each port HAProxy from C to Rust. Both passed nearly all of HAProxy's tests, but Opus 5.5 finished in 9.5 hours compared to Fable 5.1's 12 hours, and for 51% less cost.

[Mark Gerrard @MarkGerrardAI]

@evio_wwww Just tried it in a web concept bakeoff, beat gemini 3.8 flash and muse 1.3 max, Opus 5 scored them and my eye a lot better also.

Only a small test but around Opus 5 on this one for me.

[Mark Gerrard @MarkGerrardAI]

@synthwavedd I'm seeing same.

[Mark Gerrard @MarkGerrardAI]

@jebank @googlechrome Just discovered split tabs which is really handy when building a mockup to web app for fidelity.

Anyway or possible to add a scroll function that scrolls both tabs simultaneously?

[Mark Gerrard @MarkGerrardAI]

@thsottiaux @ChrisGPT Can't wait!

[Mark Gerrard @MarkGerrardAI]

@bot Awesome...

[Mark Gerrard @MarkGerrardAI]

@synthwavedd New Opus is something else.

[Mark Gerrard @MarkGerrardAI]

@trq212 Is this all cloud based so no handoff from local to cloud and back?

[Mark Gerrard @MarkGerrardAI]

@thsottiaux Perfect way to waste tokens to get to models like Fable and Astra to align with each other 😂

[Mark Gerrard @MarkGerrardAI]

@felixrieseberg How does it handle the transfer, from lid close to cloud? Is it becoming more like GrokBot?

[Mark Gerrard @MarkGerrardAI]

@bcherny It continues when you close your laptop? Is it cloud based?

[Mark Gerrard @MarkGerrardAI]

@thsottiaux @OpenRouter The Luna price cut helped, plus Anthropic abandoning a Haiku level model, Astra rocks though!

[Mark Gerrard @MarkGerrardAI]

@synthwavedd We can live without Terra.

[Mark Gerrard @MarkGerrardAI]

Built one on session.compact this week: instead of the built-in summary, a fork writes a pointer-only handoff (run ids, artefact ids, background task ids), stores it, and the transcript is replaced with it verbatim. Resumed session recalls everything. Auto-trigger still to test. One note for the team: three shapes differ from the 09-09 cheat sheet ($ calls positional, message rows use text, manifest under .claude-plugin/).

[Mark Gerrard @MarkGerrardAI]

Outside of the weekly limit change does anyone notice that Claude / Fable limits on plans seem to get used up quicker during US peak hours?

I'm UK and morning they seem to hardly move then mid afternoon onwards they burns so fast, same workloads?

[Mark Gerrard @MarkGerrardAI]

@alexsobel How is the UK going to have any sway in AI when it doesn't even have it's own working and accesible Sovereign model?

Our compute and AI is in as bad a state as our military, we're almost in the stone age compared to others.

[Mark Gerrard @MarkGerrardAI]

Since you asked sweet cheeks ☺️: 25 years software engineering and infrastructure, last 4-5 years building on frontier models daily, MD of a UK Plc that deploys them.

So I'm the deployer you'd make liable. Deployer liability already exists (negligence, Equality Act, DPA). What 'AI Assurance' adds is a mandatory audit regime — for weights I can't see, while the model vendor walks. Who audits the black box?

[Mark Gerrard @MarkGerrardAI]

@RichardBurgon How else are newer parties expected to compete with established ones?

Oh you don't want that do you?

[Mark Gerrard @MarkGerrardAI]

The Government is kicking the defence budget down the road, and doing the same with compute. Both are in a similar state, and both rest on the same assumption — that someone else will cover us. Depending on foreign models is NATO logic applied to AI. It's an economic risk before it's a security one.