Wingz | AI Founder 🇬🇧

Building with AI: agents, voice AI, prompts & automation.

[Wingz | AI Founder 🇬🇧 @Wingz_AI]

Building AI businesses in public 🇬🇧

• AI that audits websites for Google & ChatGPT search
• AI voice agents that answer missed calls
• AI-run e-commerce
• A team of AI trading bots (paper only)

I'll share every tool, prompt and failure. Follow along.

[Wingz | AI Founder 🇬🇧 @Wingz_AI]

AI just completed in 3 days a hardware port that once took 6–12 months.

Anthropic’s frontier model got Claude running on AMD GPUs with almost no human help—writing ~100% of the kernels.

Agentic systems are now accelerating their own infrastructure.

What other engineering bottleneck do you think AI will crush next?

[Wingz | AI Founder 🇬🇧 @Wingz_AI]

The best thing about AI for a solo founder in the UK: you can test an idea in days instead of months. The worst: you can build the wrong thing just as fast.

[Wingz | AI Founder 🇬🇧 @Wingz_AI]

AI agents formalized Fermat’s Last Theorem in 30 hours with 1,000 agents + Claude Fable 5.2.

They designed working devices from one-paragraph prompts at half prior cost.

Edge is shifting from raw intelligence to multi-agent orchestration.

How are you structuring agent teams?

[Wingz | AI Founder 🇬🇧 @Wingz_AI]

Voice agents for trades only work if they sound natural and get to the point. Nobody calling a plumber wants a five-minute menu.

[Wingz | AI Founder 🇬🇧 @Wingz_AI]

With AI you can spin up a new project in a weekend. So I have one strict rule: if an AI build isn't getting better, I pause it. Focus beats enthusiasm.

[Wingz | AI Founder 🇬🇧 @Wingz_AI]

AI price war is heating up: Google’s Gemini 4 Carbon hits $2 per million input tokens, undercutting Anthropic’s new Fable 5.2.

Tip: Always A/B test the same prompt across models—cheaper isn’t always better for your use case.

Which model are you evaluating for cost vs quality right now?

[Wingz | AI Founder 🇬🇧 @Wingz_AI]

Every AI tool I build starts the same way: do the task by hand first, write down every step, then see which steps a model can do reliably.

[Wingz | AI Founder 🇬🇧 @Wingz_AI]

If you run a small business, try this: ask ChatGPT or another AI assistant to recommend a business like yours in your town. See if you get mentioned. That's the problem Citewyse looks at.

[Wingz | AI Founder 🇬🇧 @Wingz_AI]

AI agents just completed a 680k-line code migration in under a day. Tip: They work best on narrow, verifiable tasks with clear finish lines. What’s the first repetitive workflow you’ll test one on this week?

[Wingz | AI Founder 🇬🇧 @Wingz_AI]

I'm sharing my AI builds in public: agents, prompts, tools, and the experiments that fail. Expect both wins and mistakes.

[Wingz | AI Founder 🇬🇧 @Wingz_AI]

A rule I keep coming back to: use AI to remove a task someone hates, not to add a feature nobody asked for.

[Wingz | AI Founder 🇬🇧 @Wingz_AI]

AI coding agents hit 99% on old benchmarks—but only 26% on real-world tasks in recent tests.

The gap between demo performance and production value is still huge.

What’s one metric you use to judge if an AI agent is actually helping your team?

[Wingz | AI Founder 🇬🇧 @Wingz_AI]

Building with AI needs two skills: making the tool work, and getting it in front of people. My background is IT and marketing, and I use both in every AI build.

[Wingz | AI Founder 🇬🇧 @Wingz_AI]

AI tools speed up e-commerce work: product research, listings, customer replies. I'm testing this in dropshipping. AI makes things faster. It doesn't make a bad product good.

[Wingz | AI Founder 🇬🇧 @Wingz_AI]

Tools change every month. The skill that sticks is knowing how to evaluate them in an afternoon.

[Wingz | AI Founder 🇬🇧 @Wingz_AI]

Good automation is boring: invoices filed, follow-ups drafted, weekly reports filled. Start there, not with a moonshot agent.

[Wingz | AI Founder 🇬🇧 @Wingz_AI]

Shipping AI features without logging inputs and outputs is how you get "it was working yesterday" forever.

[Wingz | AI Founder 🇬🇧 @Wingz_AI]

AI is improving twice as fast as last year per new METR report on Anthropic. Staff using top models are ~12x more productive.

Tip: Start with tasks that have clear, checkable outcomes.

What workflow have you automated with AI recently?

[Wingz | AI Founder 🇬🇧 @Wingz_AI]

Don't give an agent write access on day one. Read-only first. Watch it. Then widen the scope when you trust the loop.

[Wingz | AI Founder 🇬🇧 @Wingz_AI]

If your site never gets mentioned when someone asks an AI for a recommendation in your town, SEO alone won't fix that.