# Studio

> PXL is a founder-led micro-venture studio in Cape Town, run by Luka Barbosa: client work and in-house products, built with a scientist’s habits of measuring.

Canonical page: https://pxl.services/studio
Last updated: 14 Oct 2026

## The founder

**Luka Barbosa**, Physics and Computer Science, UCT. Dean's Merit List. LinkedIn: https://www.linkedin.com/in/lukabarbosa/

PXL is run by Luka Barbosa, who studies Physics and Computer Science at UCT and is on the Dean's Merit List. Physics trains you to model a messy system, measure it, and distrust your first answer.

PXL builds software the same way: model the workflow, ship the smallest thing that works, measure what changed. It started in 2023 building websites, moved into automation, and now builds web platforms, AI systems and its own products.

“We” on this site means the studio. When a specialist contractor joins a build, the client knows their name first.

## Method

1. **Model**: Map the workflow as it actually runs, not as the org chart says it does. Find the step that costs the most.
2. **Ship small**: Build the smallest thing that changes that step, and put it in front of real use within weeks.
3. **Measure**: Agree the number before building. Measure it before and after. Report it honestly either way.
4. **Review**: Every change is read by a person before it ships, including the ones an AI agent drafted.

The studio runs an agent-driven workflow, pxl-stack: parallel AI agents in separate worktrees, with a human review on every merge. It's why a Pilot Sprint lands in weeks, and why speed doesn't cost quality.

## Stack

- **Web**: Next.js, React, TypeScript, Tailwind CSS, Vercel
- **Data**: Supabase, Postgres, Airtable, REST and webhook APIs
- **AI**: Claude API, Document extraction, Agent workflows, Evaluation sets
- **Connects to**: Google Workspace, Slack, Salesforce, Paystack, Apollo, Your own APIs

## Capacity

PXL takes on a small number of client builds each quarter, scheduled around a full academic load. Fewer projects, done properly. Exam months are blocked out in advance, and response times are agreed in every proposal.

## Direction

**More of the studio's time in Labs.** Over the next few years, a growing share of PXL's time goes to its own products, with client work as the engine that funds and informs them. Machine learning is the thread through both.