Thank you / One model leaves, a new one joins / How can I help?
8 July 2026
Build With Oli
Hey,
Something a bit mad happened this week — a few thousand of you signed up in the space of a couple of days.
If you're one of them: I'm Oli, I run @buildwitholi on Instagram, daily practical stuff about building real things with AI. And at the end of this month I'm leaving my job to do this full-time, which is equal parts exciting and terrifying.
One thing up front:
I will only email when something genuinely interesting comes up, so there won't be a regular cadence to these for the time being!
So, a question before anything else. You signed up for a reason — there's something you're trying to build, or get better at, or stop getting burned by.
Hit reply and tell me what you'd like the videos and guides to cover — I read every response, and what you send decides what I make next.
[News] One model leaves, a new one joins
- Anthropic has pulled Fable 5 from paid plans — temporarily, they say; it's a capacity thing, not a strategy thing.
- In its place: Sonnet 5 — near-Opus results on coding and agent work, now the default on most plans.
The catch: Sonnet 5 talks more
- It runs on a new tokenizer that turns the same text into roughly 30% more tokens, and it spends more of them thinking out loud.
- On a paid plan, that's why your usage limit feels like it's draining faster.
- On the API it compounds — a Sonnet 5 job can genuinely cost more than the same job on Opus, despite the cheaper per-token price.
- There's a discount (about a third off, until the end of August) softening this for now. Cheap per token is not the same as cheap per job.
[Builder Resource] 100+ AI Models, Free
NVIDIA hosts over a hundred models — Llama, DeepSeek, Mistral and more — that you can use completely free. Perfect for the tasks that don't need Claude-level brains (see the staffing piece below).
Read it →[Claude User Resource] Make Your Claude Chat Last Twice as Long
Every message you send makes Claude re-read the whole conversation, which is why long chats eat your limit. Subagents give the heavy lifting its own separate memory, so your main chat stays light. Especially worth knowing now the models are chattier.
Read it →[Blog] Choosing a model is a staffing decision
You wouldn't put the senior partner on the photocopying — and you shouldn't put your biggest, priciest model on every task. My latest piece, written as a short story about a consulting firm. I think it'll change how you pick models.
Read it →Let me know what you think — and like I said, any ideas you want me to run through, just shout and I'll look to make a video of it.
Chat soon,
Oli
Want proper one-to-one coaching on this? I work with a small number of people directly. Details here, or just reply to this email.