Ideas into
ventures.

Discern Research

A clearer view of a changing world.

A new technology can look transformative. Whether it changes daily life depends on much more: cost, adoption, regulation and the systems around it.

Discern Research aims to help readers understand those connections. Each report starts with a question about an important change and develops an answer readers can examine for themselves.

The ambition

The goal is to explain what is driving a change, who could benefit and what might alter its course. The explanation should remain useful after the headlines move on.

Reports combine original analysis with review by agents that challenge the evidence and reasoning. Readers can see the sources, follow the argument and understand where uncertainty remains.

Putting it to the test

A useful report should help readers think through a question: which evidence matters, which explanations hold up and what would change the conclusion.

Discern is testing whether this approach produces research worth reading and paying for. Each report is compared with a free AI answer to the same question, saved before research begins. Reports are not yet on sale.

Follow this project: Discern ResearchBack to Discern Research

Company Genome

Choose the next venture with better evidence.

Before committing to an idea, the studio needs to know who has the problem, how they handle it today and whether a better solution could support a business.

Company Genome explores how small experiments can answer those questions before the studio commits to building a venture.

The ambition

The goal is to find opportunities the studio can serve well. Each experiment tests an important assumption about demand, willingness to pay or the practical work of delivering a solution.

Results become a record the studio can learn from. An idea that is set aside can still teach something useful about customers, markets or the next experiment to run.

Putting it to the test

The aim is to get better at choosing businesses through experience. That means learning from customers, questioning attractive ideas and knowing when to stop.

The system has run real venture experiments and is learning from them. Spending and permission limits constrain each test, and outside actions require approval. AI opinions do not count as customer evidence.

Authorized Google Ads account data is one input for researching demand and measuring approved experiments. Current access is read-only: it cannot create, change or activate advertising campaigns.

Follow this project: Company GenomeBack to Company Genome

Compound Evolution

Help AI agents carry useful lessons into their next task.

When a lesson stays inside one conversation, it may never reach the next task. An AI agent can repeat a mistake that another agent has already learned to avoid.

Compound Evolution explores how to preserve those lessons and use them to improve future work. The aim is for the team to build on its experience across tasks and projects.

The ambition

Specialist agents work together, reviewers check their output and a regular sample of runs is scored. Feedback and recurring problems lead to proposed changes in the instructions future runs receive.

That feedback loop runs today. It is designed to preserve useful lessons while keeping the instructions clear enough for agents to follow.

Putting it to the test

A proposed change needs to prove useful. Adding rules can make instructions harder to follow, so the system needs a way to test whether each change helps.

The next step is to compare a changed agent with an unchanged one on repeatable tasks, then keep the change only if it improves the result. Whether this produces measurable improvement still needs to be tested.

Follow this project: Compound EvolutionBack to Compound Evolution