refazer

Software, remade with AI

Software, remade.

We use AI to remake software — new agents where a job needs one, and the products people already depend on made ten times more useful.

See the workPurpose, vision, values

The work

Built, not demoed.

1 project

AgentData & analyticsLive

Insighter

Ask your data warehouse a question. In English.

A cross-warehouse AI analyst that answers natural-language questions over enterprise data — through a business catalogue a human approved, not a guess at what your column names mean. Every figure it reports is traced back to the rows that produced it.

Explore

The thesis

The bottleneck was never ideas.
It was the cost of building them.

01

Agents as technology partners

Not autocomplete and not a chatbot. Each one owns a real job end to end, and is accountable for the result.

02

Trust before autonomy

Software that cannot be checked cannot be trusted with anything that matters. Everything here is built so a human can see exactly what it did, and why.

03

Ten times, not ten percent

Ten times the productivity for the team that builds it, and software ten times more useful for the person who has to live with it.

Engineering

How it actually works.

All notes →

The model writes the plan, not the SQL

Insighter's analysis SQL is assembled by rules from an approved catalogue rather than authored by the language model. That choice looks conservative and is mostly about being able to sleep.

architecturesql

Why a human has to approve every metric

Letting a model infer what your columns mean is the fastest path to a demo and the slowest path to a system anyone will trust. Here is the trade we made instead, and what it costs.

architecturetrust

Want to see one of these working?

The fastest way in is a live walkthrough against real data. Tell us what you are trying to fix and we will show you whether we can help.

Get in touch