STRATEGYOngoing practice85% confidence

The Augmentation Company

Always solve tasks just a little too hard for the model — keeping a human in the loop is what builds the data flywheel.

Problem it solves

How an applied-AI company should position relative to full autonomy so that capability compounds.

Best for

Applied-AI companies choosing between an augmentation product and a fully autonomous "lights-off" solution.

Not ideal for

Mature, fully-solved tasks where the model already exceeds humans and oversight adds only cost.

Overview

Why this framework exists

Adept's strategy is to always be an augmentation company, not to deliver a lights-off solution. Framing yourself as augmentation keeps you working on tasks just a little too hard for today's model, so a human stays in the loop providing oversight, clarification and feedback — and that is exactly what builds the data flywheel and forces you to develop core AI capabilities faster than a lights-off competitor. The people who do the jobs become supervisors and troubleshooters who parallelize work and click into any agent trajectory. Because the reward signal comes from real customers, evaluations are concrete customer needs — not academic or simulator evals — which Luan believes lets you build AGI faster.

Core principles

3 total
  1. Stay an augmentation company: always work just past the model frontier so a human stays in the loop.
  2. Human oversight on too-hard tasks is the data flywheel, not a temporary crutch.
  3. A real-customer reward signal makes evals concrete and compounds capability faster than academic evals.

Origin story

How this framework came to be

Adept's enterprise rollouts, and Luan's answer to "why not build a general foundation-model lab" from guests Kanjun Qiu (Imbue) and Raza Habib (Humanloop).

Source

Traced to primary
Source · PODCAST
Why Google failed to make GPT-3 + why Multimodal Agents are the path to AGI — with David Luan of Adept
Latent Space (swyx & Alessio) · 2024
Open source →

Related frameworks

Browse all Strategy →