How We Build AI Systems: Process and Method

The process

First call to live system

A free 30-minute call. We find the highest-value problem to solve first.
Architecture, deliverables and the numbers that mean success, agreed in writing.
Progress you can see every week, tested against real edge cases.
Handover with documents and full access. An optional retainer as models and APIs change.

The system map

Follow one message through a live system

This is Inbox Triage, the system that handles our own enquiries. Pick a path to see which steps a message takes, or select any step to see what happens there.

The contact form checks the message and queues it. The team is emailed a copy at the same moment, so nothing depends on the queue alone.
Every message becomes one job. A job in progress is claimed, so two workers cannot both handle it, and the queue stops taking work at 500 open jobs.
Fixed rules drop bulk mail and anything from a no-reply, bounce or internal address, and flag urgent words and support problems.
Bulk mail and robot mailboxes stop here. Our own domains and no-reply, bounce and notification addresses are refused before anything is sent.
The model reads the message as data and returns a category, an urgency from 0 to 5 and a "needs a person" flag. If the call fails, the rules decide alone.
The AI's labels can only make the outcome more careful. Complaints, support problems and urgent messages always wait for a person.
A routine enquiry gets a short, fixed acknowledgement. The system does not write its own replies; only your first name is used.
Everything that is not routine sits on a review board until someone approves a reply or closes it.
The worker sends from the outbox. A failed send is retried after 1, 2, then 4 minutes and so on, and set aside for a person if it keeps failing.
The result of every send is recorded. Finished messages are deleted after a set period; the details are in the privacy notice.

The method, and a system taken apart

Read how it works before you buy

Method

The Project Factory

Twelve phases from brief to audited release. Research comes before the stack, the architecture comes before any AI agent, and ten audit checks are run by code.

See the 12 phases Worked example

Inbox Triage, taken apart

The system that handles our own enquiries, step by step: the queue, the rules, the AI labels, the policy gate, the review board and the outbox.

See how it works

How the studio runs

One reviewed place that people and agents work from

Most small studios run on scattered notes, chats and tools, and AI agents inherit that mess. We keep one governed home for how the business thinks and operates, the SlickLab Command Center, so people and agents draw on the same reviewed knowledge. It is private: we share the idea here, not the build.

Knowledge

Reviewed before trusted

Nothing counts as approved knowledge until a person has checked it.

Agents

Agents with guardrails

Automation works inside written rules, with a record of what it did.

Secrets

Secrets stay out

Credentials never live in the knowledge itself.

The standard

Four things every system has to do

1

The rules decide

AI reads, labels and drafts. Fixed rules in ordinary code decide what is allowed to happen next.

2

A person can step in

Anything urgent, sensitive or unclear is handed to a person, and a customer can always ask for one.

3

It says what it is

Anything that talks to your customers introduces itself as an agent, never as a person, and contacts only people who agreed to be contacted.

4

Every step is recorded

Each action leaves a record you can read back, and the same request arriving twice is handled once.

See the products → Draft a project brief Start a project

How we build AI systems you can check