Shola Oladipupo

I build AI agents and the evals that keep them honest. Solutions architect in London. Built ResumeRadar (8K+ scans) and WorkCrew. Views my own.

[Shola Oladipupo @sholabuilds]

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.

[Shola Oladipupo @sholabuilds]

@omarsar0 The gap is that nobody tests what the agent decided to remember. on one-off tasks it doesn't matter, on domain work a wrong memory just keeps coming back

[Shola Oladipupo @sholabuilds]

@emollick The craft moves rather than disappears. with agents most of my time now goes on deciding what it's allowed to do and checking what it did

[Shola Oladipupo @sholabuilds]

@googleafrica "the people best placed to build AI for Africa already live here" 👏🏾 congrats to all 10 teams, can't wait to see what they ship

[Shola Oladipupo @sholabuilds]

@omarsar0 own computer and browser is the interesting part. curious how it asks before it does something on a site you can't undo 👀

[Shola Oladipupo @sholabuilds]

@HamelHusain error bars should be the default on every eval chart. a 2 point gap inside the noise isn't a result 📊

[Shola Oladipupo @sholabuilds]

@dabit3 bookmarked. managing agents is starting to look like managing a team: clear scope, a clean handoff, and someone checking the work

[Shola Oladipupo @sholabuilds]

@swyx 👀

[Shola Oladipupo @sholabuilds]

@TokataHQ @lanfrica @UmuIgboUnite @myabiadaily @Voiceofigbos Chris and NaijaVoices are doing important work 👏🏾 speech recognition that properly handles Igbo would open up so much

[Shola Oladipupo @sholabuilds]

@tessl_io @lamismukta @AnthropicAI @_lopopolo "context can multiply model intelligence" is a great way to put it. would love to see the slides if they get shared @lamismukta

[Shola Oladipupo @sholabuilds]

@ClaudeDevs Running a full app inside Claude changes what you need to test. Once app tools can act from inside the chat, the question isn't just whether the call works but what the model does when the tool returns something unexpected. How are permissions scoped per app tool?

[Shola Oladipupo @sholabuilds]

@swyx @lucasmeijer @FabianHedin @BuiltByJustin @kentcdodds @aiDotEngineer rehearsal energy looks great 🔥 enjoy AIE Code

[Shola Oladipupo @sholabuilds]

@claudeai Whiteboards and Sheets in the same place is a nice combo for planning a build 👀

[Shola Oladipupo @sholabuilds]

@paw_lean @rhiannon_io @OpenAIDevs Cambridge on Wednesday 👀 any London ones coming up soon?

[Shola Oladipupo @sholabuilds]

@omarsar0 @i14labs fixed-size KV cache at 1M tokens is the part I'd want to test. does recall hold up for stuff buried in the middle, or just the ends?

[Shola Oladipupo @sholabuilds]

@Teetolan This is great 🙌🏾 I did the Claude certs this year, nice to see people in Naija getting sponsored for them. Is it open to non-devs too?

[Shola Oladipupo @sholabuilds]

@alexalbert__ the da Vinci one is so good 🤯 history lessons are about to look very different

[Shola Oladipupo @sholabuilds]

@dabit3 I'd add a fourth: a clear point where it stops and asks. In my multi-agent sample, if routing confidence is low it asks the customer a clarifying question instead of guessing

[Shola Oladipupo @sholabuilds]

@ericzakariasson @bot 👀 bring it on

[Shola Oladipupo @sholabuilds]

@simonw "This codebase is yours" is such a simple fix 😂 going to try it on the next ResumeRadar refactor

[Shola Oladipupo @sholabuilds]

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.

[Claude @claudeai]

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.