Twelve months ago we ran our first AI agent inside a live client account. Today there are agents in roughly two-thirds of our retainers, doing measurable work on bid management, creative iteration, content drafts, lead scoring and weekly reporting. That's not a pitch. That's an operations report.
What we've learned over those twelve months is mostly the opposite of what the LinkedIn essays said it would be. The biggest wins came from the boring agents nobody talks about. The biggest wastes were the agents we were certain would print money. And the operators who got the most out of this stuff weren't the ones who moved fastest. They were the ones who were honest about which decisions their team could actually own.
This is what we'd tell you over a coffee, with twelve months of P&L behind it.
The agents that pulled their weight
Three categories earned their seat at the table. They share one thing in common: high repetition, low judgement.
Paid search bid and creative agents
Roughly eighty per cent of our agent value sits here. An agent watching a Google Ads account every fifteen minutes, with the right guardrails, will catch wasted budget faster than a senior strategist reviewing the account weekly. We typically see eight to fourteen per cent improvement in CAC inside the first sixty days, with no change in spend. The agent doesn't get bored on a Friday afternoon.
Content drafting agents tied to a real brief
When a senior strategist owns the brief and the brand voice file, an agent can ship first drafts of SEO articles, category pages, ad variants and email flows at four to six times the pace of a human writer, at comparable quality after one editing pass. Note the conditional: tied to a real brief. An agent without a brief produces fluent nonsense at scale.
Reporting agents that read the dashboards so we don't have to
The unsung win. A reporting agent that reads GA4, Search Console, Meta Ads Manager, Shopify and the ad accounts, then produces a one-paragraph commercial brief per channel every Monday morning, frees up roughly a day of senior time per account per month. Multiply that across a book of twenty accounts and the maths gets interesting.
The agents we thought would win, and didn't
This is where we paid the tuition. Three things we backed that didn't earn their keep:
- Lead-qualification agents wired into inbound forms. They handled the easy seventy per cent of cases (which the existing form fields handled fine anyway) and got the difficult thirty per cent wrong in ways the client noticed.
- "Strategy" agents that read the data and recommended the next move. They generated plausible-sounding briefs that occasionally contradicted what the data actually said. The senior strategist still has to do that work. The agent helps her gather inputs, not form the view.
- Always-on social listening agents. More noise than signal. The signal was already on the Search Console dashboard. We were just paying twice for it.
The pattern is what you'd expect, but boring once you see it: agents are excellent at compressing the work, less good at owning the judgement.
Operators who got the most from agents weren't the ones who moved fastest. They were the ones who were honest about which decisions their team could actually own.
What we tell every operator who asks if they should "be doing AI"
Three things, in order. We've had this conversation roughly forty times in the last six months, and the answer keeps narrowing to these:
- Pick the most repetitive, lowest-judgement piece of work in your marketing stack. Almost always: bid management, creative variants, or weekly reporting. Start there. Skip the strategy agent. Skip the customer-service bot. Both will burn you in the first ninety days.
- Buy or contract a senior human who's run agents in production. The cost of getting this wrong is mostly opportunity cost (six months of a campaign you should have been running properly) and brand cost (a customer interaction that didn't sound like you). A senior pair of eyes makes both go away.
- Decide upfront whether you want to own the agent or rent it. Both work. Renting is faster to launch and easier to start. Owning is more expensive on day one and pays back when the playbook is yours, not the agency's. We've published the two paths in detail: rent if you want to test the value, own if you've decided this is core capability.
The honest read: the line between an agency that uses AI well and one that doesn't is going to harden in the next twelve months. The clients who are getting agents right are the ones treating them as junior labour with a senior on top, not as a magic shortcut around hiring or expertise. The agents pay for themselves. The senior judgement on top is what stops them from costing you ten times what they saved.
If you want our read on what an agent setup would look like in your business, drop your URL into the free audit and we'll send the read back inside thirty seconds.