# Questions people ask (/docs/about/faq)

---
type: Document
title: Questions people ask
description: Direct answers on what an FDE does, what a connector is, whether the 70% is real, what happens at a context reset, and why the internals are not published.
status: stable
order: 5
generated: { by: "process:claude-code", at: 2026-08-27T09:00:00Z }
verified:
  - { by: "human:rrizwan98", at: 2026-08-28T14:00:00Z }
ksor:
  audience: [public]
  owner: human:rrizwan98
  approval: { by: "human:rrizwan98", at: 2026-08-27T09:30:00Z }
trust_tier: human-reviewed
build_id: sha256:d9f989bee8fa4c15cf0f7f1572f9e9a18305ef4e2ce0c3cfb6a5a421ead6dc1d
dirty: true
---

## What does a Forward Deployed Engineer do?

Works inside the customer's real workflow, connects the AI to the tools the
team already uses, deploys it, and stays accountable for the result in
production. Not a demo handed over.

## AI already writes code. What is left to automate?

The typing was automated; the delivery was not. A person still writes every
instruction, re-explains the project after every context reset, redoes the
rework, and remembers to run the browser test, fix the deploy and write the
handover. Those four jobs are where the systems go. See
[how I work](how-i-work.md).

## Is "70% automated" a real number?

It is counted per run, not estimated. A website-delivery run has ten steps and
three of them are human decisions, so seven of ten need no person. The table is
on [how I work](how-i-work.md). The count for another workflow is made the same
way during the audit and agreed before the build.

## What is an MCP connector?

A remote server that ChatGPT, Claude or Codex attaches to, carrying one
company's process and rules. The server enforces the steps, so any AI that
connects behaves like a trained employee of that company. Explained with an
example on [what I build](what-i-build.md).

## What happens when the AI's context resets?

Nothing is lost. Every build has an identifier and its state lives in a
database rather than in the conversation, so a new session resumes from the
last completed step instead of restarting the project.

## Can a whole company run this way?

Yes: work enters from the CRM, AI staff do the jobs, one person reviews and
approves. It starts as one workflow, not as a company-wide programme. The shape
and an example are on [what I build](what-i-build.md).

## Will the AI take actions without approval?

High-impact actions keep a human gate, and the exact approval points are agreed
before deployment. In a typical run the human steps are the brief, the design
approval and the go-live.

## Why do you not describe the systems you have built?

Because they are clients' and a company's, not this record's to publish. What
each live connector does and refuses is public; the architecture, prompts,
agent names and client names are not. The concepts are explained in full with
examples instead, and the rest is discussed in the audit.

## Can you work with our existing tools?

That is the default. The connectors run from ChatGPT, Claude Code or Codex, and
the workflow is designed around the tools, permissions and constraints the team
already has.

## How is our data protected?

Each part of a system gets only the access its job needs. Data paths,
permissions and logs are reviewed as part of the design, and secrets never live
in the code.

## What happens when a system fails?

Failure paths are designed before launch. The system can retry safely, stop for
review, or return the work for correction; the run stays reviewable either way.

## Who maintains it afterwards?

The deployment includes documentation and handover. Ongoing optimization
reviews failures, improves the evaluations, and extends only what proved
useful.

## How do we start?

With an audit of one workflow. Bring the job where a person still babysits the
AI: see [contact](contact.md).