I build AI agents and the evals that keep them honest. Solutions architect in London. Built ResumeRadar (8K+ scans) and WorkCrew. Views my own.
London, UK
Joined October 2017
- Tweets1.9K
- Following1.6K
- Followers1.3K
- Likes1.2K
Pinned Tweet
I'm Shola. Solutions architect in London, and I build AI agents on my own time.
What I've shipped:
• ResumeRadar: free CV checker, 7,000+ scans
• WorkCrew: AI automation for UK small businesses, replacing WhatsApp and spreadsheet chaos with systems that run themselves
• A multi-agent customer-service sample with human handoff, published through AWS samples
Here I post what worked, what broke, and the evals that told me which was which.
Follow along if you build with Codex, Claude Code, Cursor or Grok Bot.
6
13
537
Anthropic just announced Claude Fable 5.2 and Mythos 5.2 👀🔥
Exciting. But before I swap a new model into anything real, it goes through my own test cases first. The ugly ones. Wrong tool, vague question, someone asking for data they shouldn't see.
If it passes those, then it's an upgrade.
Introducing Claude Fable 5.2 and Claude Mythos 5.2, designed to tackle the most daunting technical challenges.
They’re capable collaborators on ambitious projects in many fields, and show improvements in areas like coding, reasoning, and vision.
1
3
63
My degree is in agricultural economics, from Ibadan. Now I'm a solutions architect building AI agents.
What got me across was a 6-month secondment while I was a data analyst at NHS England. I earned 4 certs and did real cloud work in that window.
Moving into tech from another field? The fastest route is often inside the job you already have. Ask for the project before the new title.
1
1
68
Whenever I can use plain code instead of the model, I do.
In my multi-agent sample, the supervisor routes with a keyword scorer, not an LLM. If the best score is under 0.34, it doesn't guess. It asks a clarifying question.
Determinism is a feature. Save the model for the parts that are actually fuzzy.
1
3
49
I think the human handoff is the most important feature in a customer-service agent, not the fallback.
In my multi-agent sample, I built the escalation path before I tuned a single answer. When the agent can't help, the customer goes to a person with the context attached, so they never repeat themselves.
If you can't say exactly when your agent hands over to a human, it isn't ready for customers.
3
2
47
Spent yesterday at @OpenAI DevDay Exchange in London. The best part were about how building itself is changing.
4 things I'm taking away 🧵
2
3
139
3. ChatGPT Sites started as one engineer's internal prototype. He was tired of juggling separate accounts for the repo, hosting, database and auth. It spread inside the company and became a product.
The best tools still start with your own frustration.
1
1
8
At OpenAI DevDay Exchange in London today. Packed room.
My biggest takeaway so far isn't a model. It's plugins and MCP. Agents can now subscribe to events and act on them, not just wait for you to ask.
I think that makes one question more urgent than ever: when an agent acts on an event by itself, what does it check before it acts?
@OpenAI @wulfie_bain_ @paw_lean
1
5
124
I've earned 13 certifications. I think none of them taught me as much as shipping one small tool that strangers actually use.
ResumeRadar has now scanned 7,700+ CVs. Real users find the problems no exam ever asks about.
If you're collecting certs, ship one small thing alongside them. The certs make more sense afterwards.
2
1
52
In case you're wondering what this means for you:
Benchmarks are moving from "can it pass an exam" to "can it help with my actual week".
If you build agents, steal the idea. Ask your users what they'd actually hand over to an assistant, and test against that list, not a leaderboard.
For AGI to benefit all of humanity, it should help people with the everyday tasks that matter most to them. So we built LifeVal, a new benchmark shaped by feedback from thousands of ChatGPT users on where they most want help from a personal assistant—from finding childcare to accessing government services to managing personal finances.
Over the past year, we’ve used LifeVal to measure and improve how well our models handle these everyday tasks.
openai.com/index/lifeval/
42
In case you're wondering what this means for you:
On the $100 plan, nothing changes but the name. It's just "Max" now.
On the $200 plan, now "Ultra", the 5-hour cap goes once your account is verified.
Weekly limits stay the same. So you can run longer sessions, but the total you get each week hasn't grown. Pace your big runs across the week.
We’re renaming our Max plans:
Max 5x → Max
Max 20x → Ultra
Max is still $100/month, and Ultra is still $200. If you’re a subscriber, there’s nothing you need to do.
We’re also removing the 5-hour limit for verified Ultra accounts, starting today with most existing subscribers. Newer accounts keep a temporary limit until they’re verified, which helps us prevent fraud and abuse. You'll get an email when it's removed.
Weekly limits remain unchanged. View our plans here: claude.com/pricing
57
If your agent reads big PDFs or runs long computer-use sessions, you were probably hitting the 32 MB cap well before the context window ran out.
Now you get about 4x more room before you have to split the job up.
We've raised the max request size for Messages and Token Counting on the Claude API from 32 MB to 128 MB.
Image-heavy requests like long computer-use sessions and big PDFs were previously hitting 32 MB with plenty of context window left. Now you get ~4x more room.
2
2
62
Open weights at the top of FrontierCode for open models is a big one for teams that can't send their code to a hosted API 👀
Today, we release the weights of a new and improved Le Chonk, advancing the open frontier for agentic coding and cybersecurity.
💻 Built to build: Le Chonk ranks #1 on @cognition’s FrontierCode among open models, and sets a new benchmark for Pareto efficiency on @artificialanalysis.
🛡 Built to defend: State-of-the-art in cyberdefense, put through the paces by the world's leading cyber vendors, including @CrowdStrike, @cloudflare, @rubrik, @zscaler, and others.
Built for sovereignty: Make it yours. Download the weights on @huggingface , customize it with Mistral Forge, and deploy it wherever you need it.
Read more from our team and customers 👇:
40
In case you're wondering what this means for you:
If you're on Pro 100, 200 or 500 or Business Premium, your usage limits double from today. Pro 200 isn't being reduced as planned, so Work and Codex usage stays the same.
If you've been rationing long agent runs, now's the time to plan them again.
Day 25/
We are doubling all usage limits for all Pro 100/200/500 tiers and Business Premium, effective today.
We heard your kind feedback and won’t be reducing usage limits for Pro 200, so your Work and Codex usage will stay the same. We’re also going to continue offering extended limits for dots for now, and usage of your dot will still continue to be separate from your default usage.
And we'll of course also keep shipping better models and other efficiency improvements that increase your effective usage over time.
1
38