How Cognition AI Does Forward Deployed Engineering

These are my notes from the Forward Deployed Engineering workshop at @aiDotEngineer on June 30, 2026.

Session: How Forward Deployed Engineering Is Done at @cognition
Speaker: Jia Wu
Company: Cognition AI
Company X: @cognition

Problem Space

Cognition's framing: FDE maximizes the overlap between product capabilities and customer problems.

In software development, writing code is about 20% of the problem. Testing, reviewing, and maintaining are the rest.

Deploying agents without direction wastes tokens and spend. FDE identifies high-leverage strategic initiatives and automates agents to run continuously.

Hiring Philosophy

Cognition looks for a T-shaped persona:

• Broad client, business, and process skills.
• Deep technical spikes.

They hire from diverse backgrounds, including:

• PMs.
• Founders.
• Software engineers.

Technical depth and domain expertise are harder to teach than business sense.

Core mantras:

• Relentless curiosity.
• Customer obsession.

Metrics and Case Studies

Cognition shipped about 10 times more high-quality PRs over six months using its own agents.

A three-month customer embed delivered about 150% additional headcount equivalent. Delivery timelines compressed by about 82%. PR volume nearly doubled compared with engineers using single-point tools.

Philosophy Shift

The early era focused on maximizing tokens with little budget pressure. The current shift is toward measurable delivery outcomes.

The differentiator is making the whole organization faster, not just individual engineers. Field insights de-risk the roadmap.

My Take

Cognition's framing makes FDE feel less like a services layer and more like a strategy layer for agent deployment.

The core lesson is that agents need direction. Running agents without a high-leverage target wastes tokens and review attention. The FDE finds the strategic initiative, shapes the workflow, and makes the whole organization faster, not just one engineer faster.