From agency to operator: 6× output, almost 2× margin, no new senior hires.

The client is a Czech-headquartered performance marketing agency, anonymized here. Forty-seven people across Prague and Bratislava, founded in 2016, working with mid-market consumer brands across CEE: paid search and social, programmatic display, creative production, conversion optimization, marketing automation. They were profitable, well-respected by their clients, and stuck.

Their bottleneck was the same one every services business hits at 40–50 people. Senior strategists were spending 70% of their time on production work — writing ad copy, briefing designers, building reporting decks, optimizing bids, drafting client emails — instead of strategy. The next 10 hires would not improve margin; they would just spread the same work across more people. The agency's CEO had read the standard "AI for marketing" content and was skeptical of all of it. The standard pitches were either glorified prompt libraries (worth perhaps a 5% productivity bump) or end-to-end SaaS platforms that locked the agency into a vendor's roadmap and pricing.

He came to us with a specific brief: "I want my strategists to operate Claude Code the same way they operate Photoshop. I want it on the desk. I want it in their hands. I do not want a SaaS subscription that we rent. I want a capability that we own."

Sixteen weeks later, the agency runs seven production agents that every team uses daily, has built four more internally, and has tripled the number of campaigns it can run per strategist. They have not raised retainer prices — they have simply taken on more clients with the same headcount. Margin is up, salaries are up, and the partners are deciding whether to stay closely-held or scale aggressively.

The constraint: agencies fail at AI when they treat it as a tool, not a discipline

The agency had already tried the obvious thing. They had a "ChatGPT subscription for everyone" policy in 2023. After six months it had produced exactly two material productivity improvements, both of them by the same one strategist who had spontaneously taught herself good prompting practices.

The lesson the partners drew was correct: a chat interface in a browser tab is not a workflow. It is a curiosity. To change the agency's economics, the AI had to be inside the production process: in the same file system as the campaign briefs, the same repository as the asset templates, the same Slack channels as the team handoffs. It had to be operated like a craft tool, not consulted like a search engine.

That is what Claude Code is. It runs in the strategist's terminal, against the agency's own files, with access to the agency's own tools, integrated into the agency's actual production pipeline.

The portfolio: seven production agents covering the campaign lifecycle

Wave 1 — Built by Orchestrary in months 1–4

Wave 2 — Built by the agency's own team using Claude Code (months 5–9)

The aggregate impact: an agency with operator-class economics

Senior strategist time on production work: 70% → 22%. Senior strategist time on strategy and client thinking: 20% → 55%. Time-to-launch on a new campaign: 9 days → 3 days. New business meeting hit rate: 18% → 34%.

Strategic outcome

The agency's partners are now in the position every services-business owner dreams of: their unit economics no longer require linear hiring to scale. The next 50% of growth will come from raising productivity per head, not adding heads. They are deciding whether to stay closely-held with a high partner profit share, or to scale aggressively and challenge the regional Big-Four-style media agencies — a conversation that simply was not on the table 12 months ago.

Why "deploy AND teach with Claude Code" was right for an agency

1. The work is creative; the operator must be in the loop

Marketing campaigns require taste judgments that no agent in 2026 makes well unsupervised. Claude Code keeps the strategist on the keyboard at the moment of creative decision — the agent is producing options, not shipping campaigns. A SaaS platform that automated the work end-to-end would have produced uniform, cheap output indistinguishable from any other agency's.

2. The agency wanted to own the leverage

A SaaS subscription would have shifted the agency's productivity gain to the SaaS vendor's pricing. Within 18 months, the vendor would have raised prices to capture most of the surplus. Building on Claude Code, with the agency's own files, prompts, tools, and configurations, means the productivity gain stays with the agency.

3. The Academy turned a creative team into builders

The agency's 22 Academy graduates included copywriters, art directors, account managers, and performance specialists — almost none with traditional engineering backgrounds. By the end of Track 3, two copywriters were writing tools (in Python, with Claude Code's help) that called external APIs.

The human dimension

"I have been running this agency for ten years. Last year was the first time I felt like the partners were running a business instead of just being expensive operators. I have my brain back. So do my strategists."

Founder & CEO

"I used to write briefs for two days and resent every minute of it. Now I edit a draft for forty minutes and spend the rest of the day on the actual interesting question — what should this brand be doing in the next quarter. It is not just faster. The work itself is better, because I am not exhausted by the time I get to the strategy part."

Senior Strategist · Wave 1 user

"I built MEETING-PREPPER over three weekends in the Academy. The weeks I started using it, I noticed I was actually awake during my client meetings. I am managing twelve accounts now. Before, I would have drowned at six."

Account Manager · Wave 2 builder

"The Academy taught me to write Python. Six months later I co-built one of the agency's most-used agents. My career is in a place I would not have predicted 12 months ago."

Junior Copywriter · Wave 2 co-builder

What we did not do

• Build agents that ship campaigns autonomously. No agent in this engagement publishes a single ad, sends a single client email, or moves a single budget without a human approval.
• Replace the strategist with the agent. Every agent is positioned as "you with more time," not "you replaced."
• Pitch a SaaS spin-off. The agency's partners briefly explored productizing the agent stack into a SaaS for other agencies. We advised against it.

The deliverable

The contracted commitment was nine months. We exited at month 9. The agency now runs seven Wave-1 agents and four Wave-2 agents, all owned and operated by their own team. The next four agents on the roadmap are entirely theirs.

"You sold us a sixteen-week engagement. You actually sold us a different agency."

Founder & CEO · Engagement closeout