Micro Agency. The category just got named — and it’s the shape we’ve been building toward.

A few weeks ago Rize published something called the Micro Agency Playbook. Five tenets covering people, tools, positioning, pricing, and visibility. The playbook reads less like a thesis and more like a checklist someone wrote after watching the same shape emerge across half a dozen companies that don’t otherwise have anything in common. Every ships fifteen people, five products, and a consulting business. Crosby prices contracts by the document, not the hour. Pace publishes Agent Operating Procedures. Butler/Till runs media buys against a published media-cost-reduction target. The Forward Deployed Engineer pattern Pragmatic Engineer wrote up describes small pods doing what large teams used to. None of these companies set out to be a “Micro Agency.” They just kept making the same architectural choices.

This is a piece about a category that just got a name. If you run a consumer brand or a software company wondering whether your next hire should be a person, a tool, or a partner, the Micro Agency category tells you which kind of partner is actually new — and which kind is the same retainer with AI in the deck. Leanboat is a Micro Agency for consumer and software brands. That’s the manifesto sentence, and the rest of this piece is the receipts.

The traditional agency model only works because nobody has stopped to ask whether scale should still cost what it costs. The Micro Agency category is what the answer looks like once you do.

The five tenets, with the named exemplar behind each

Fewer people. Every ships five products and a consulting business with fifteen people, and the subscription business is reported at seven-figure ARR. The number isn’t a constraint; it’s the design choice.

Better tools. The stack is the new headcount. Per-client documented workflows, agent-orchestrated execution, real-time visibility into hours-per-deliverable. The named term varies — Compound Engineering at Every, Agent Operating Procedures at Pace and Decagon, our own Codified Engagement at Leanboat — but the shape is the same. A documented operating layer that the operator and the agents both read.

Sharper positioning. Crosby doesn’t sell “legal services.” It sells contract turnaround at a published median time per document. Specificity is the moat. The category-defining work is choosing the one job to be famous for.

Outcome pricing. Crosby prices by the document. Butler/Till anchors against a published media-cost-reduction target. The hour stops being the unit.

Real-time visibility. The hours, the agents, the cost-per-deliverable — visible to the operator and to the client, on demand. Not a quarterly report. Pace ships its Agent Operating Procedures publicly because the visibility is the product as much as the output is.

What this looks like at Leanboat scale

Two founders. Per-client Codified Engagement file — a CLAUDE.md knowledge base of the brand, the customer, and the channel; agent workflows tuned to that knowledge; approval gates where human judgment lives; a kill switch when something goes sideways. The “team” the client gets is small. The system the team runs on is large.

At the budget tier we serve, the canonical three-role marketing chart often collapses to one operator plus the right tools. That’s a current view, not a fixed rule — but it’s worth naming because most “AI-native agency” pitches don’t address it. They describe enterprise-team-shapes to small-budget buyers and then quietly bill the same way the old retainer billed.

What gets installed in the first month, in five stages.

Plan. Humans set strategy — which channels, what voice, what the rubric looks like. The plan is the rubric the agent runs against.

Generate. The system produces variations against that rubric: captions, layouts, briefs, hooks, multiple options per slot.

Review. The operator approves or rejects. Named gate. Nothing publishes without an operator pass.

Publish. The system ships and schedules. No human moves files between tools.

Measure. What performed feeds the next plan. The loop closes monthly. The rubric sharpens every cycle.

One workflow, fully installed

The clearest example I can show you is the one this post comes out of: how Leanboat runs its own content research.

The problem. Two founders. A daily publishing cadence on a fast-moving topic where the voice has to be operator-real, not outsourced or templated. A normal agency setup would either drop the cadence, dilute the voice, or burn founder hours writing prose every day. None of those was acceptable.

The install, mapped to the five stages above.

Plan. I set the umbrella topic, the pillar mix (operator-voice pieces alternate with analyst-voice pieces, no more than two consecutive same-pillar), and a content bench — each candidate piece tagged with the third-party voice it engages, the counter-position it answers, and the signal it’s testing.

Generate. A scheduled morning step refills a two-draft buffer. It picks the next pillar that satisfies the streak rule, pulls source notes and counter-references for that piece, and writes a draft skeleton against a rubric of failure modes I’ve spelled out by name.

Review. I spend about an hour on the draft. Approve, edit, or kill. Most posts ship within that hour. Analyst pieces that earn it get a protected three-hour block once a week.

Publish. The system commits the post, deploys the site, and writes what shipped — title, surfaces, timestamp — into the per-engagement file so the loop closes on a record, not memory. No moving files between tools.

Measure. Calls booked carry the most weight, then citations of the post by AI-search products, then engagement from named operators in the field. The signal feeds a weekly review on cadence and a monthly learning pass on whether the topic mix is still working.

What’s not in the picture: a productivity-tool stack glued together by hope. The agent has access to a per-engagement file that knows the topic, the voice rules, the counter-position ledger, and the failure modes I refuse to ship past. That file is the Codified Engagement. Everything else runs against it.

That five-stage shape is what we install for TheNextGuide’s Tour Studio — direct integration with the operator’s tour catalog, not photo-send-and-pray — and the same shape applies to a B2B SaaS daily research brief, a DTC retention flow, an AI-search content cadence. The function is what changes. The loop is constant.

What this is not: a productivity-tool reseller, an “AI center of excellence” inside a holdco, or a generic AI agency that promises capability and ships chat-window output. The Micro Agency framing carries weight because it commits to a shape, not a feature list.

Where it goes wrong

“Micro Agency” as rebranding without the architecture. The risk in any newly-named category is that incumbents pick up the label and keep the old shape. The five tenets are the discipline that makes the label load-bearing. Outcome pricing is the litmus test — if the category is real, the pricing changes. If the category is a rebrand, the retainer survives.

The “we use AI” agency pitch. Most agencies bolting GenAI onto retainer work fail the test for AI-native services that the venture investors who underwrite this space are now publishing. Revenue and logo retention don’t count; the proof is AI doing a material share of the work at high gross margin. Micro Agency is the operator-side version of that thesis. If the work doesn’t compound from one engagement to the next, the architecture isn’t there yet.

That is what a Micro Agency for consumer and software brands looks like.

Book a strategy call →