Why I Stopped Worrying About AI Eating My Agency

Two years ago, I was nervous. Every LinkedIn post screamed that agencies were dead. Every podcast guest claimed solo founders with AI agents would replace teams of forty inside a year. Today I run Otto Media with around 42 clients across SEO, digital PR, web design and Google Ads. So I've had a front-row seat to what AI does, what it breaks, and what it leaves alone.

Here's the boring truth nobody says cleanly. AI does not change agencies the way the hype machines insist. What it does is reorganize margin, compress execution, and pulverize the mediocre middle while rewarding specificity and trust at the edges. The dying agency is the 50-person generalist averaging 8% net margin, selling integrated marketing services to an SMB that just bought ChatGPT Pro. The thriving agency is small, specialized, founder-led, and built for taste over volume.

Let me unpack what I've learned across three areas: operations, planning, and leadership.

Operations: the stack matters less than the workflow

Every founder asks me about my AI stack. So here's the honest answer. The tools are commodity. We use Claude for writing, Surfer for SEO optimization, n8n for automation, Otter for meetings, Jasper for brand voice consistency, and AgencyAnalytics for white-label reporting. Total cost sits under $500 per month for the core stack. Almost any agency could copy it tomorrow.

Then comes the real question, which is what you do with the workflow. One US agency, Xponent21, documented 4,162% organic traffic growth in twelve months by producing over 100 AI-assisted SEO pieces under a clear editorial framework. Another shop dropped reporting time from 30 hours per month down to 10. These wins did not come from better tools. They came from disciplined process design and ruthless quality control.

Now here is where most agencies fail. They turn AI loose without guardrails and produce slop. Macquarie Dictionary made "slop" its 2025 Word of the Year for a reason. Google's March 2024 spam update wiped out sites running scaled, low-value AI content, and some lost a million monthly visits overnight. Meanwhile Billion Dollar Boy found consumer preference for AI-generated creator content fell from 60% to 26% in two years. So quality has to be a workflow input, not a hopeful output.

For me, every AI draft passes through a human editor before it touches a client deliverable. Sources get checked. Each client brand voice gets a custom GPT trained on their existing copy. That discipline is the difference between scaling and self-destructing.

Planning: pricing is the real AI question

If you're still billing by the hour, AI just cut your revenue in half. Karl Sakas put it bluntly earlier this year. What used to take 12 hours now takes 4. Under hourly billing, that's a two-thirds revenue loss for the same outcome.

So we shifted Otto Media to value-based and productized pricing about eighteen months ago. Some services run as fixed-fee monthly retainers tied to outcomes. Others sit in productized tiers with clear scope and turnaround. The conversation with clients changed too. Now we anchor on the outcome we deliver, not the time we spend. When a client asks why a website costs the same despite faster delivery, the answer is simple. They invested in tools and expertise to get results faster, and the price reflects the result, not the hours.

Then there's the moat question. Because AI commoditizes execution, the moat has to live somewhere else. Five places work for creative services. Vertical specialization, where David C. Baker's rule of thumb is 10 to 200 competitors and 2,000 to 10,000 prospects. Proprietary workflows packaged as IP. Founder personal brand and trust capital. Community and audience. And taste, which AI cannot synthesize and which Rick Rubin built an entire Anthropic project around.

Plus, the small-but-mighty thesis is empirically real. Brett Williams runs DesignJoy at $2 to $3M ARR solo. Justin Welsh has crossed $10M cumulative at 89% margin with one part-time VA. Marc Lou cleared $1M in 2025 with no team. These aren't outliers anymore. They are the new ceiling for what one operator with a niche and a content engine can build.

Leadership: taste is the whole job now

Here's what I keep coming back to. When AI handles execution, leadership becomes the entire product. McKinsey said it well in their 2025 leadership report. AI cannot set aspirations, make tough calls, build trust, or generate genuinely new ideas. So the founder's job compresses into three things. Frame the brief. Hold the creative line when AI produces 80% generic output. Own the client relationship through trust capital.

That last one is underrated. Edelman's 2025 Trust Barometer shows institutional trust has collapsed, but trust in individual businesses has held. Founder-led agencies inherit that. AI can help clients find you, but it will not make them trust you. And trust is what gets you onto the shortlist.

Then comes the burnout trap. AI gives leverage, but leverage without control is the fastest route to founder collapse. Justin Welsh wrote about his panic attack from running 150 reports as a VP of Sales, and how it felt like Tetris speeding up until the pieces stack. Most founder burnout in 2026 looks like that. Too many AI-assisted projects open at once. Too many tabs. Too many outputs to review. And not enough of the deep judgment work the AI can never do.

So I built rhythms instead of relying on willpower. Cal Newport's daily shutdown ritual takes ten minutes and clears the rumination. Greg McKeown's "if it's not a hell yes, it's a no" filter kills 80% of the optionality AI generates. And the music keeps me sane, because specific knowledge feels like play to you but looks like work to others. For me, that's still the decks and the live sax.

The real bet for 2026

So here's my contrarian wager. Boring authenticity beats polished AI every time. And 42 clients served by 8 people with sharp taste will out-compound 420 clients served by 80 people with mediocre processes.

Neither view is mainstream consensus yet. Both will be by 2027.

What AI cannot do, in the end, is decide what's worth doing. Pricing your taste, holding the creative line, telling clients the truth they don't want to hear, sequencing the work like a DJ reads a room. That remains the job. The agencies that survive 2026 will be the ones whose founders treat AI as the world's best junior employee with no judgment and no accountability, and themselves as the only person in the building who has both.