About António

António Costa is co-founder of NewData.Tech and a seasoned data leader who has served as fractional or full-time head of data at companies including OLX Group and Arrive Group. He helps scale-ups mature fragmented data estates into AI-ready platforms — not just warehouses full of tables, but governed, contextualized assets that automated agents can actually consume.

What we talked about

AI readiness versus data volume — Why most companies have tons of data but rarely data that is AI-ready — and what that readiness layer actually requires.

The semantic layer as secret sauce — How business meaning, KPIs, table relationships, and domain language get encoded so AI tools understand what your data actually represents.

Four pillars of AI-ready data — Metadata capture, governance, security, and reusability — plus why AI amplifies existing mess instead of forgiving it.

Startups versus incumbents — Smaller companies move faster with workarounds, but scaling demands the same governance layers big marketplaces already built over decades.

From SQL to natural language — A case study from Arrive Group where business users query data in plain language and spin up dashboards in minutes instead of waiting on analysts.

Augmentation over replacement — Why António believes every role will evolve into orchestrating agents — and companies that cut headcount instead of multiplying output are missing the point.

Sharpest moments

On the gap between having data and using it:

rarely it is AI ready.

On what makes the semantic layer essential:

the semantic layer is the secret. It's actually the secret sauce.

On missing context:

without the right context, AI can only infer or invent. And then it can be a mess.

On why messy data gets worse with AI:

AI will only amplify the mess that you have in your data layer.

On the shift from human-facing to AI-facing data:

Now it's different. Now it's making that data understandable and right for AI.

On the future of work:

whatever role you are at, you will need to orchestrate agents.

Key takeaways

• Having lots of data is not the same as having AI-ready data — context, semantics, and governance are the real unlock.
• The semantic layer translates business tables, KPIs, and domain terms so agents query accurately instead of inventing answers.
• Metadata, governance, security, and single-source-of-truth reusability are the four pillars António keeps returning to.
• GDPR and data protection are real concerns for maturing European scale-ups — not something to hand-wave away.
• Data accumulated over years is the hardest moat to replicate — even when frontier models are broadly accessible.
• Natural-language data access is finally fulfilling the old SQL promise — but only on top of well-structured foundations.
• António is firmly on the augmentation side: orchestrate agents, produce 10x more, don't fire the humans.