Manifesto

01Who we are

Rémy comes from ENSAE, and works on interpretability and on systems that hold at scale. Lucas comes from École polytechnique and the Corps des Mines, and is fascinated by the phenomenon of emergence, in living systems and artificial ones, and the ways each explains the other. We have run projects outside of school for as long as we have been in it.

The venture-backed scalable-SaaS path was open to us. We want to solve concrete problems, learn with our hands, and carry social responsibility inside the company we build. Sagace is a step towards that decision. We are two today, and we will only work with people who want to learn this field from inside a real problem.

02Our project

We bring machine learning into scientific organizations that make consequential decisions currently without it, starting with biotech.

Four examples, and the list is not exhaustive. We design experiments, framing active learning and lead optimisation mathematically and applying them to the next experiment a team runs. We build first machine-learning systems for scientists, on their own data and on the public datasets they have not opened yet. We produce evidence artifacts: toxicity assessments for competitive intelligence, technical reports substantive enough to carry part of a fundraise. And we want to do research that truly matters, on open questions we take on with a partner and that we can ideally publish.

Three projects we could concretely start tomorrow. We would design the linker of an antibody-drug conjugate alongside a chemist, through optimal experiment design. We would assess the toxicity of a drug already on the market and produce the report that funds a safer one. We would extend Jack Scannell’s work on pharmaceutical productivity, in a decade where AI changes the cost structure of discovery.

Our ideal client is a biotech with many processes and many decisions, and no machine learning informing any of them.

03What we believe about AI

AI should be cooperative. Most of the industry is building the opposite.

Full autonomy is the cheap engineering path for AI systems. Systems that keep a human inside the loop are harder to design, and yet they are achievable. When the human becomes the bottleneck for an AI system, the system was designed without them.

The same economic gains are reachable on both the autonomous and the cooperative paths. One of them leaves people with work, judgement and authorship, while the other removes them. This is a design choice, made now, by people who are free to choose otherwise. If autonomous systems win on ease of construction alone, AI-governed organizations will compete with human-governed ones across most sectors, and they will win.

We reject universal basic income as the answer to all this. People want purpose, accomplishment, and the right to have ideas that count. The human characteristics of work deserve protection as fundamental rights.

Capability has outrun application. The distance between the frontier labs and the people who could use these tools is the most wasteful gap in the field, and it is where we work.

04How we work

We work in six-week sprints with a hackathon in the middle. We work full remote, with regular checkpoints under review and autonomy between them. Every project closes with a confidential deliverable, an open-source contribution wherever the work allows one, a case study, and a proposal for what to do next.

We take on people newer to the field than we are and review everything they produce, because learning is part of our deliverable. We invest in the ecosystem around us through events, introductions and collaborations. People who work with us should leave with better careers than they arrived with.

We would refuse work we already know inconclusive, even if it is profitable. We refuse engagements we cannot walk away from. When a client is wrong, we tell them, because their budget has a better use, with us or elsewhere.

05A promise

We publish our failures whenever confidentiality allows. We price so that a small team can afford us. We work with founders whose ambitions for the public good match our own.

06Today

Our budget is small. We work on our own time and ask no more of our collaborators. We take no venture capital, which caps how fast we grow, and we accept that ceiling. Another route exists: a company that everyone around it wants to keep working with does not fade away.

We want AI to be useful. We want to solve interesting problems. We are starting with the decisions scientists make.