Published, free to read

The Agency AI
Infrastructure Stack.

Five layers behind consistent AI work for your agency’s clients, in the order they have to exist.

Start with what the work needs

Why the order matters.

Your team may already have useful AI setups: an assistant with a client’s files, a workflow for preparing reports, or instructions for drafting their content.

A gap can appear when that client changes an offer. One colleague updates the assistant; another starts the next draft from a saved copy of the old brief. The question is whether the confirmed change reaches the setups where the next task happens.

Access and workflows matter. They also depend on something less visible: a current, agreed version of the information behind the work.

Here are five layers to check for one client your team works with regularly. Each gets a job, a picture of what can go wrong, and a question to answer with the account owner. Use an actual client brief and the current sources, and keep that client’s context separate from the others.

Build from the source outward.Each layer uses the decisions made before it.
  1. 01
    Shared context

    Agree what is current.

  2. 02
    Access & environments

    Decide who can use it, and where.

  3. 03
    Skills & repeatable work

    Define how a recurring task runs.

  4. 04
    Governance

    Own, review and communicate changes.

  5. 05
    Automation

    Run suitable, tested preparation.

The order is a build guide. Context, access and ownership still need attention after a workflow is running.

Layer 01 / 05

Shared context

The job
One current, agreed version of what is true about a brand — positioning, audience, offer, messaging, working rules — for everyone and every AI environment that touches the work.
What can break without it
Two colleagues can be drafting for the same client from different briefs. One includes a retired service; the other has the current offer. If the correction stays in a chat, the old service can return in the next draft.
The question for your agency
What would AI read about this client before drafting their service page, and who confirmed that the offer, audience and claims are current?

Constructed example · not a client

Meet Rasterwerk.

Rasterwerk is invented for this example: fourteen people in Hamburg, mostly working with industrial B2B clients. Here, its team is drafting a service page for a fictional industrial maintenance company. Every detail below is part of the illustration.

The client’s website still mentions emergency call-outs. The current brief is for planned maintenance. An old sales deck promises round-the-clock cover.

Before drafting the page, Rasterwerk needs the client to confirm the current offer and who it is for. That decision belongs in the shared client context, where both the account team and their AI sessions can find it.

Rasterwerk / client contextConstructed sample
Current client offer
Planned maintenance for industrial pumping systems. Emergency call-outs are outside the agreed offer.
Who the page is for
Maintenance managers in manufacturing plants who are responsible for keeping those systems in service.
What a draft must make clear
Which systems the service covers, what happens during a maintenance visit and which client source supports any performance claim.
Still open
Whether food-production sites are in scope. Rasterwerk’s account lead needs the client to confirm that before naming it as a served sector.

The open question tells the person drafting the service page which decision is still outstanding. AI gets the same boundary. The reviewer can check the draft against the confirmed offer and send an unresolved question back to the account owner.

If your tools change: one approved current version per brand still has to exist, and people and AI environments have to read the same one.

Back to contents

Layer 02 / 05

Access and environments

The job
Decide who works in which approved environment with which material, and make those boundaries visible in the setup.
What can break without it
A useful client setup may live on one colleague's laptop. If another person cannot access the approved context, they may start from an old copy. When the account changes hands, it can be unclear which environments still hold the client's material.
The question for your agency
Which approved environments hold this client's briefs and brand context, who can use them, and who removes access when someone leaves the account?

If your tools change: access is granted per brand and revoked in one place, and no individual holds credentials the agency cannot see or replace.

Back to contents

Layer 03 / 05

Skills and repeatable work

The job
Turn recurring work into a defined task with a named input, a defined output and an owner, so colleagues can run and improve the same task.
What can break without it
A colleague may have useful instructions for drafting a client's monthly report. If those stay in a private chat, a new account team has to reconstruct them. Improvements made for one report may never reach the next person preparing it.
The question for your agency
For a recurring client task, such as a monthly report or service-page draft, can a colleague find the current instructions, required sources and review steps?

If your tools change: recurring work belongs in a shared, versioned definition the team runs from, not in one person's memory.

Back to contents

Layer 04 / 05

Governance

The job
Set who may change the shared material, who reviews it, and who owns the current version.
What can break without it
If a client changes their positioning and only one colleague updates their working material, an old description can still reach the next draft. Without a named owner for the change, a reviewer may also be checking against the old version.
The question for your agency
When this client changes an offer, who confirms the change, updates the shared context and checks that it reaches the people and AI environments doing the work?

If your tools change: someone owns the current version, and the right to change it stays narrower than the right to work with it.

Back to contents

Layer 05 / 05

Automation

The job
Take repetitive preparation off people, once the four layers underneath hold.
What can break without it
If a workflow runs before the source information is agreed, each run can carry an old offer or an unverified claim into another draft. More output also means more places for the team to check.
The question for your agency
Which recurring preparation already runs automatically for this client, which version of their context does it use, and who reviews the output?

If your tools change: automation is what you add once the foundation holds, and it earns a task only when the task clearly recurs and its output is already good.

Back to contents

Put the questions to work

Which layer needs
your attention?

Take the five questions to the person responsible for this client’s work. Write down what exists, where it lives and who keeps it current. Mark the unanswered questions as gaps, then repeat the check separately for the next client.

You can buy tools and use an existing structure for the system. Your agency still has to decide what is true about the brands it works on, who has access and who approves changes.

Start with the first gap that prevents the next layer from working. If people disagree about the offer, agree on the source before automating the draft. If the source is current but one colleague cannot use it, look at access and loading.

The next useful step is the one that gives the person doing the work a better starting point.

When you are ready to build

Two ways to put
the foundation in place.

Build it with your team

Blueprint €99

Follow the manual and templates to set up shared context, access, connections and Skills in your own accounts. Your team makes the decisions and does the build.

See the Blueprint

Have me build the agreed setup

Install €3,990

I install the shared infrastructure in your agency’s accounts and hand it over to your team. You provide the information, access and decisions. The Blueprint is included.

See what gets installed

Net prices, one-time payment. Tax is calculated at checkout. All sales are final. Read the offer for its scope, access and purchase terms.

Christoph Sauerborn

Christoph SauerbornAI Systems Engineer for Agencies. I install the shared brand and technical foundation behind how agencies use AI.