Announcing our $8m seed round from Afore Capital, Picus Capital, Focal VC, HF0, and others
Subliminal Labs
Request a demo
Careers at Subliminal

Build the system that understands how companies work.

Subliminal observes how work actually happens, reconstructs the workflows behind it, and builds automations validated against real work. We’re a small team of engineers and researchers working on the hardest part of AI adoption: giving models the operational context of a specific business.

See open roles Get in touch
  1. Observes how work happens.

    Gather operational data and action logs across the organization, capturing the context behind how tasks are completed.

  2. Finds what can be improved.

    Identify recurring workflows, inefficiencies, and the automation opportunities with the greatest business impact.

  3. Builds and deploys automations.

    Create automations matched to business needs and current model capabilities, testing them against thousands of instances of real work.

  4. Keeps improving.

    As models advance and the business evolves, update existing automations and deploy new ones.

Backed by$8m seed
Afore CapitalPicus CapitalFocal VCHF0
Read the announcement
Why this problem

The bottleneck in AI adoption is context, not capability.

Models are already capable of automating far more than businesses have put them to use. The bottleneck is not capability. It is context: the people who can build effective AI rarely know how a specific company operates, and the people who know the company rarely build AI.

Today that gap is bridged by hand. Forward-deployed engineers embed inside a company, interview employees, shadow their work, and sometimes do the job themselves. It is slow, disruptive, and captures a fraction of what happens, and even that goes stale as the business changes.

We think an AI-native company needs comprehensive, continuous operational understanding instead. That is the system we are building, and it touches everything from native desktop capture to LLM pipelines to automations running inside real companies.

See what that looks like at Fella Health
What you’ll work on

One system, from the desktop to the automation.

Native capture clients

A Swift menu-bar client for macOS and a Go system-tray client for Windows record desktop activity after sign-in and OS permissions, then ship evidence to regional ingestion APIs. Updates go out silently through signed update channels.

Workflow reconstruction

Pipelines that turn hours of recorded work into task units, then into reconstructed workflows with their steps, tools, inputs, outputs, and exceptions, each linked back to the original evidence.

Automations platform

Agents built around a team’s real workflows and tested against completed work before they run, with outcomes, pausing, and human review in the dashboard.

Regional data infrastructure

Multi-region storage and processing lanes on AWS, including deployments inside a customer’s own account, managed with Terraform and synchronized to a global control plane.

Dashboards and evidence tooling

A React dashboard where customers explore workflows, people, and the evidence behind them, plus internal review tools for inspecting capture sessions and pipeline output.

Evaluation harnesses

Recording harnesses, fake applications, and backtests that measure whether each stage of the analysis engine does what its design says, on real recordings rather than fixtures.

How we work

Written down, proven on real data, shipped the same day.

  • Design docs are the source of truth.

    Every subsystem starts with a written design. Code that contradicts the doc is a bug, not a new convention.

  • Proof over assertion.

    Nothing is done until it has been shown working end to end: real hardware for the clients, real recordings for the pipeline, real rows in the database.

  • Ship continuously.

    A merge to main deploys the global API and dashboards within minutes. Client releases reach employees through the same silent update channels they already use.

  • Privacy is a product constraint.

    Subliminal is not an employee monitoring tool. It learns how work happens, not how productive each person is, and private data is discarded.

  • A small team owns the whole system.

    Each project has its own toolchain and owner, and you will work across the stack rather than inside one layer of it.

  • Everything is checked against real work.

    Workflows, automations, and the analysis engine itself are validated against thousands of instances of work employees actually completed.

Who you’ll work with

Former quants, AI researchers, and engineers from OpenAI, MIT, Stanford, Jump Trading, and Optiver.

  • Matthieu Huss

    Matthieu Huss

    MIT researcher and post-training startup founder. Won largest AI/Hardware competitions in China and USA.

  • Tomer Moran

    Tomer Moran

    Built and operated critical software for hospitals; responsible for platform, security, and compliance.

  • Ozaner Hansha

    Ozaner Hansha

    Previously internal AI infrastructure, agentic capabilities, and data partitioning at Optiver.

  • Andrew Mackenzie

    Andrew Mackenzie

    High-performance computing, deep learning research, and AI infrastructure at Jump Trading. Previously Amazon ads.

  • Tyler Killian

    Tyler Killian

    Built and maintained internet-scale data crawlers at Exa. Putnam top 500.

  • Ethan Chang

    Ethan Chang

    MIT researcher in AI for robotics. Previously Apple and OpenAI, evaluations for first reasoning models.

More about the team
Open roles

We hire in small numbers.

Every hire changes what we can build. We look for people who have shipped hard things end to end and want to own a whole problem, not one layer of it.

Current openings

Roles are listed on our hiring site, along with what each team is working on right now.

See open roles

Don’t see your role?

If you have built something hard and want to work on this problem, tell us what you built and what you want to work on next. Links to your work matter more than a résumé.

Email [email protected]

Come build it with us.

Tell us what you’ve built and what you want to work on next.

Get in touch