We went to Opticon NYC this year as an Optimizely partner, which meant two days in the room instead of a recap read afterward. This year's event is branded "Camp Opticon," with sessions sorted into tracks called The Trailhead, The Bonfire, The Outpost, and Tent Talks, like someone crossed an enterprise roadmap with a Boy Scout jamboree. Every core product got the word "Agentic" stapled to the front of it too: Agentic CMS, Agentic EXP, Agentic Commerce, Agentic Feature Management. At some point a roadmap stops being a roadmap and starts being a find-and-replace.

Underneath the camp cosplay and the buzzword bingo, though, session after session made the same real point: the busywork is over. Agents will structure your content, tag it, publish it, run your tests, flag your features, and adjust in real time. One roadmap talk promised, "agents that test, learn, and adjust in real time." Another pitched agents that handle the busywork of structuring, tagging, and publishing content so teams can focus on strategy instead. A third framed the whole shift in six words: "AI Runs the Tests. You Run the Strategy."

Sit through enough of that back-to-back, and a pattern falls out that's bigger than any one feature. Optimizely is making a bet about what a marketer's job is actually for. A few sessions later, on the same day, I sat in on a talk called "Humanity Is the Premium Layer." Its thesis, straight from the abstract: AI creates abundance. Human judgment creates value.

The busywork was never the job.

Nobody built a career in marketing because they loved tagging content or babysitting a test until it hit significance. Nobody's LinkedIn headline says "Passionate About QA-ing UTM Parameters." That was always the toll you paid to get to the actual work: understanding a customer well enough to say something true to them, at the right moment, in a way that made them trust you a little more than they did yesterday.

Optimizely's own numbers make the point better than we could. One session referenced 154,000 experiments run through their platform and asked what "winning" even looks like once AI is choosing which ones to run. Another, "The Last Click Lies," made the case that the conversion you can measure was never where the decision actually happened. It happened earlier, somewhere nobody was tracking, built out of trust that accumulated over time.

AI is about to absorb the mechanical half of the job: the structuring, the tagging, the test-running, the fortieth landing page tweak of the quarter. Good. That's the plan working, and it's the same bet we've made in our own work: strip out the grunt work between a good idea and a live experience, so the time goes to the parts that actually move revenue instead of the parts that just prove you were busy.

Table stakes isn't a strategy.

When every brand has access to the same agents, the same test velocity, the same content-at-scale, none of that is a differentiator anymore. It's table stakes. Optimizely's own speakers seemed to know this. A talk titled "The Human Is Obsolete. Long Live the Human" was built to argue itself down. "He's Just Not That Into You" made the case that over-personalization is a trust killer and restraint is what actually wins. And a Bonfire discussion just asked flatly: who are you really marketing to, humans or AI?

There's something a little unhinged about an enterprise software company hosting a debate, on its own stage, over whether enterprise software is making people obsolete. Still, you don't program a room full of talks arguing both sides of the same question unless the industry genuinely doesn't know the answer yet. The shape of it is clear enough, though. When the mechanical work gets cheap, judgment gets expensive, and so does taste, restraint, and knowing your customer well enough to leave something out.

The thing that doesn't scale

We'd push further than the sessions did. If judgment is the premium layer, its highest form is finding the actual people who are your people, the specific, non-average humans who'll buy from you five times and tell three friends, and building something around them that a competitor with the same AI stack can't copy in a weekend.

Agents can run more ad variants and tighten a bid faster than any person could. None of that gets you the customer who showed up because a friend already mentioned you in a group text. That's not a targeting problem. A referral code doesn't buy the same thing as a real recommendation, no matter how well the automation behind it runs. The brand people bring their friends to without being asked wasn't built by an agent, and it's the one acquisition channel that doesn't get more expensive every time a competitor turns on the same AI.

This pattern shows up everywhere the mechanical work is getting automated out of marketing right now, including in the tools we build. Taking manual editing off someone's plate is just the mechanism. What matters is where that freed-up time goes: better creative, a sharper point of view, an actual relationship with the people you're trying to reach.

Do something real with the time back.

Three moves, if you actually run marketing instead of just attending conferences about it:

  1. Audit where your team's hours actually go this month: not what's on the roadmap, what people are doing right now. If a meaningful chunk is mechanical (formatting, tagging, manual QA, repetitive test setup), that's the first thing to hand off, not the last.
  2. Put the freed-up time somewhere AI can't follow: a customer advisory group, a community you moderate yourself, a segment you actually talk to instead of just watching in a dashboard.
  3. Get suspicious of anything that scales too easily. If a competitor with the same tools could copy it in a weekend, that's overhead you haven't automated yet, not a differentiator.

None of that needs a lanyard or a free Hydro Flask.

AI will keep getting better at the parts of this job that were always closer to administration than marketing. Good. Let it. The value was never in that work; it was in what you did with the time it freed up. Opticon didn't quite ask that question out loud, but every other session was circling it. Your AI can write the copy. It can even test forty versions of it before lunch. It still can't find your tribe for you, and it definitely can't build one.