How Agencies Actually Build AI Into Their Operations (Not Just Train On It)

Quick reminder — Boardroom fall cohort applications are still open. If you're at $1M–$3M ARR and want a room of real peers at the same stage, applications are open. Short form. We review every one.

The first five founding members get three things the members after them won't:

→       A say in how the Boardroom is structured — format, cadence, focus areas

→       A free AI Visibility Audit for their agency — where your brand shows up (or doesn't) inside ChatGPT, Perplexity, Gemini, and Google AI Overviews

→       Quarterly 1:1 roadmap sessions with me, separate from group sessions

Here's a pattern I see in almost every agency that reaches out to us after doing some kind of internal AI push:

They ran a workshop. Or they bought team subscriptions to Claude and ChatGPT. Or they hired someone to come in and teach prompting. Maybe they did all three. The team was engaged. People were interested. There was a lot of energy in the room.

Then nothing changed.

Six weeks later, a few people are using AI on their own when they remember to. The rest have gone back to doing things the way they always did. The founder is still the biggest source of output quality in the building. The workflows look exactly the same as they did before the training.

This isn't a motivation problem. It's not even a skills problem. It's a systems design problem.

Training teaches people how to use a tool. Building installs the tool into how the work actually gets done — so it runs whether or not anyone remembers to use it. Those are completely different interventions, and confusing one for the other is why most agency AI adoption stalls.

The distinction is simple but consequential:

Most agency AI initiatives live in the training column. The ones that actually change how an agency operates live in the building column.

The question isn't "does your team know how to use AI?" It's "is AI embedded in the workflows your team runs every day?" Those are different questions with very different answers.

This is the actual sequence for making AI stick inside an agency workflow — not as a training exercise, but as a permanent operational change. It applies to any workflow in any layer of the agency.

STEP 1  Document the workflow before you touch the AI

You cannot AI-native a workflow you haven't mapped. Before adding anything, write out the current steps: who does what, with what inputs, in what order, to produce what output. This takes 20–30 minutes per workflow. Most agencies skip it because it feels slow — and then wonder why the AI implementation doesn't stick.

A workflow map doesn't need to be elaborate. A simple list works:

Example — Client monthly status update workflow (current state):

1.  Account manager pulls metrics from Google Analytics and the social dashboard (30 min)

2.  Writes the update narrative in Google Docs from scratch (45 min)

3.  Sends draft to senior lead for review (1–2 days turnaround)

4.  Incorporates feedback (20 min)

5.  Sends to client via email

Total time: 95+ minutes of work + 1–2 days elapsed. Highly variable quality across account managers.

That's your baseline. Now you have something specific to improve — not "the agency" but this one workflow, with this one output, taking this specific amount of time.

STEP 2  Identify the AI insertion points — not the whole workflow

Within the documented workflow, find the specific steps where AI can handle execution while a human handles direction and quality control. Not "use AI for this workflow" but the precise moment where the handoff happens.

Looking at the example above, the insertion points are obvious:

→       Step 1 (data pull): Can't be AI-handled yet — requires logging into tools. Can be streamlined with reporting automation, but that's separate from the AI installation.

→       Step 2 (writing the narrative): This is the AI insertion point. A human pastes the pulled metrics into a Claude Project built for this client, and AI writes the first-draft narrative in the agency's voice, calibrated to this client's preferences and history.

→       Step 3 (review): Still human — but now they're reviewing a first draft that's 80% right, not writing from blank.

→       Step 4 (feedback incorporation): Another potential insertion point — paste the feedback into Claude and get a revised draft.

The result: the same workflow now takes 20–25 minutes of human time instead of 95. The quality is more consistent because the context is standardized in the Claude Project, not in the account manager's head.

STEP 3  Build the system — not the habit

This is the most important step and the one most agencies miss. The goal is not to create a new habit your team has to maintain. It's to make using AI the path of least resistance — and not using it the thing that requires extra effort.

The wrong way:

Build a great prompt in someone's personal Claude account and share it in Slack. Hope people remember to use it.

The right way:

Build the prompt into a shared Claude Project that the whole team can access. Make "open the [Client Name] project and paste the metrics" the documented step in the SOP. Now it's part of how the job gets done — not an add-on people might use.

The practical infrastructure:

→       One shared Claude Project per major client — brand voice, past examples, feedback patterns, common requests, quality benchmarks all loaded in

→       The prompt is saved in the project, not in someone's head

→       The SOP for the workflow references the Claude Project explicitly: "At step 2, open the [Client Name] project and paste the metrics report"

→       New team members inherit the full context immediately — no knowledge transfer required

STEP 4  Install the quality standard — not just the tool

The last thing missing from most AI implementations: a clear definition of what done looks like. When is an AI-generated first draft good enough to move to review? When does it need to go back for another pass? Without a calibration point, team members either over-rely on AI output (sending things that aren't ready) or under-trust it (rewriting everything from scratch).

For each workflow you AI-native, define and document:

→       What a strong AI output looks like for this specific workflow (one example is worth more than a paragraph of description)

→       The three most common ways AI output goes wrong for this workflow — and what to do when it does

→       The review standard: what the human reviewer is specifically looking for, not a generic quality check

Put these in the Claude Project itself as reference material. Now team members have a calibration point that travels with the tool — not something they have to remember from a training session six months ago.

Not a roadmap. Not a plan. One workflow, fully installed, before you read the next issue.

Here's how to pick the right one:

The most common first installation for agencies: the client status update or monthly report. High frequency, significant team time, defined output, not founder-dependent. If you have one of those, start there.

This week's task:

Day 1: Pick the workflow. Document the current steps in a simple list.

Day 2: Identify the AI insertion points. Mark which steps AI handles vs. which the human handles.

Day 3: Build the Claude Project and the prompt. Test it with real data.

Day 4–5: Define the quality standard. Run one team member through it.

Day 6–7: Roll to the full team. Review what needs adjusting.

One workflow installed properly is worth more than ten workflows half-done. The goal this week is one.

Before you can install AI into a workflow, you need to know where it goes. This prompt takes a workflow description and returns the insertion points, the recommended Claude Project structure, and a draft quality standard — the first three outputs of the installation process, ready in 10 minutes.

Run this for any workflow you're considering. The output gives you your installation roadmap for that specific workflow in under 15 minutes — the insertion points, the Claude Project spec, the prompt draft, and the quality standard. From there it's a build, not a design problem.

Running a business is hard. Building a website shouldn't be.

With Readdy.ai, just describe your business, the AI handles your website design, performance, SEO optimization, and more. 

No coding or design skills needed. Launch in 2 minutes. Your site could be live before you finish your coffee.

Try for Free.

This week's picks are all directly relevant to the build side of AI adoption — tools for creating workflow automations, AI agents, and connected knowledge systems that run without people having to remember to use them.

Voiceflow  voiceflow.com

Visual builder for AI agents and conversational workflows — no code required. For agencies building AI-native client-facing tools (chatbots for client portals, intake agents, FAQ bots) or internal workflow agents, Voiceflow is the most polished no-code option available. Used by teams at brands like JP Morgan and Walmart, but accessible enough for a 5-person agency ops team. Free tier available; paid from $50/month.

Dust  dust.tt

Context-aware AI that connects to your workspace tools — Notion, Slack, Google Drive, GitHub, Confluence — and makes that institutional knowledge available to AI agents without copy-pasting. Think of it as Claude Projects but connected to your entire knowledge base automatically. Particularly useful for Layer 1 (Intelligence) workflows where the AI needs deep company and client context to produce useful output. Built for teams, not individuals.

Bardeen  bardeen.ai

Browser automation with AI that can run research tasks, scrape web data, and move information between tools without developer involvement. For PR agencies doing media monitoring, journalist research, or competitive intelligence, Bardeen can automate the data-gathering steps that currently eat Layer 1 time. Works as a Chrome extension; AI component identifies what data to pull and how to structure it. Free tier available.

Zapier Interfaces  zapier.com/interfaces

Build custom internal tools and portals on top of Zapier automations — without code. Separate from Zapier's automation builder, Interfaces lets you create forms, dashboards, and client-facing tools that trigger automations in the background. Useful for agencies that want to build a structured intake portal, a client request form, or an internal ops tool without hiring a developer. Included in Zapier paid plans.

Browse AI  browse.ai

AI-powered web scraping and monitoring that turns any website into a structured data feed — no code needed. Train it on a web page by pointing and clicking; it extracts and monitors the data you care about on a schedule. For PR agencies tracking media coverage, competitor activity, journalist beats, or client mentions across sites that don't have APIs, Browse AI is one of the cleanest no-code solutions available. Free tier: 50 credits/month; paid from $49/month.

📊 BY THE NUMBERS

One workflow per month.

That's the installation cadence that produces the fastest results without overwhelming the team. Agencies that try to install AI across all workflows simultaneously almost always stall — too many changes at once, inconsistent rollout, team resistance, no clear wins to point to.

Agencies that commit to one fully installed workflow per month — properly documented, properly built, properly quality-standardized — have six AI-native workflows running by month six. The compounding effect of that becomes visible around month three: the founder starts having capacity they haven't had in years, account managers stop asking for approval on things they now handle independently, and client satisfaction scores tend to go up because consistency improves.

📣 BEFORE YOU GO

If this issue resonated and you want to map your specific workflows against the installation framework before committing to a build — the Agency AI Assessment is where that conversation starts. We look at what you're running, where the highest-leverage insertion points are, and what the full adoption engagement would look like for your agency. [agencyownerlab.com/ai-assessment →]

And the AI Value Audit — the free self-assessment from Issue 011 — is still available if you want to score your agency against the 3-layer framework before the Assessment conversation: [HERE]

TL;DR: Fall cohort applications now open for Boardroom, our version of a mastermind. Real peers, real accountability, and direct advisory access at a stage of business where most owners say that's the hardest thing to find. For PR + marketing agency owners generating $1-3M ARR. [APPLY FOR FOUNDING MEMBER OFFER]

See you next week.

Work smart. Enjoy life harder.

Founder, Agency Owner Lab

When you're ready, here's how we can work together:

→ The Boardroom — Get in the right room. for PR + marketing agency founders generating $1-3M ARR. Advisory, peer accountability, exclusive partner offers/resources. Applications now open for Fall cohort! [Apply here]

→ Agency AI Adoption Assessment + Engagement — Custom AI strategy for your agency. Plus option to add 3 months of fractional ops support to make sure adoption sticks. [Apply here]

→ Agency Growth Roadmap — Operational audit + systems strategy. [Apply here]

→ Founder Advisory — Your advisor. Your business partner. For the founder who uses AI for strategy but wants a real human to strategize with. Quarterly commitments. [Apply Here]

→ Implementation Sprints — done-for-you systems builds, Standalone or paired with another program. [Book a Systems Audit]

→ Agency OS Lab (Community Membership) — SOPs, Claude installs, tool stacks. $97/month. [Join the waitlist here]