The execution crew is now software.
Klaviyo moved Composer into public beta at the start of July. You type a prompt — “build a win-back campaign for lapsed VIPs with a 20% incentive” — and it hands back a launch-ready campaign: audience segment, copy, send timing, channel split, the lot. Everywhere I saw it discussed, the reflex reaction was the same question: is this the thing that replaces us?
I’ve read everything Klaviyo has published about it, and I think that’s the wrong question. The right question is older and more interesting: what happens to a craft when the equipment stops requiring a crew?
What Composer actually is
The facts first, because most of the commentary skips them. Composer was announced in March 2026 as a private beta and opened to public beta at the start of July. It’s an agent that takes a plain-language brief and generates complete campaigns and flows inside your Klaviyo account: it picks who to target, writes the messaging, decides how many touches, chooses between email and SMS, and sets timing. It can also run the other direction — auditing your live flows and campaigns and handing you a ranked list of what’s underperforming and what it would change.
Two details matter more than the demo videos. First, the outputs are grounded in Klaviyo’s aggregate performance data: their pitch is fourteen years of results across 193,000+ brands, so the agent is drawing on what has actually worked at scale rather than generic best-practice folklore. Second, nothing sends without human approval. Composer drafts; a person ships. Klaviyo’s co-CEO Andrew Bialecki framed the launch bluntly: “The execution layer in software is moving from humans to agents.” Note which layer he named. Execution, not judgment.
Klaviyo’s own projection is that a team running 3 campaigns a month could run 15. Hold that claim; I’ll come back to it, because whether it’s a promise or a threat depends entirely on who’s approving the sends.
The camera stopped needing a crew once before
Around 2008, Canon released a stills camera that happened to shoot cinema-grade video. Footage that previously required a rental house, a focus puller, and a lighting truck could suddenly come out of a body that cost less than a used car. Filmmakers argued about whether this was the death of the craft.
It wasn’t. What died was the requirement that a film crew exist before a film could. The volume of footage in the world exploded, and almost all of it was bad — because the crew was never what made films good. The directors with taste and judgment got dramatically more powerful: one person could now shoot what used to take eight, and the scarce skill shifted from operating the equipment to knowing what was worth shooting and recognising, in the moment, whether the take was any good.
Composer is that camera, for retention marketing. The execution crew a proper email program used to require — a copywriter, someone building segments, someone assembling templates, someone scheduling and picking channels — collapses into the tool. What doesn’t collapse is the director’s chair: the brief, the standards, and the final call on whether this take ships or gets re-shot.
What the agent can’t know about your account
Here’s where practitioner scepticism earns its keep. Composer’s intelligence is aggregate: patterns from nearly 200,000 brands. That’s genuinely valuable — it’s why its drafts will beat a junior marketer’s on structure and timing most days. But averages are exactly what they sound like, and the failures I get hired to fix are almost never average.
Composer will build a beautiful abandoned-checkout campaign on top of a Placed Order metric that your payment provider never actually fires. It will draft a re-engagement push for a segment whose “engagement” is mostly Apple’s machine-generated opens. It will happily target “all subscribers” on an account whose list is 60% dead weight, because it can’t smell the deliverability damage the way an operator who has watched a domain reputation crater can. The agent inherits your account’s data quality completely and uncritically. Clean accounts will get scary-good drafts. Messy accounts will get confident, well-formatted extrapolations of the mess.
Which brings back the 3-to-15 claim. Fifteen campaigns a month is not a goal; it’s a capability. On a disciplined account with tight engaged segments, more well-targeted sends are more revenue. On an undisciplined one, quintupling volume is the fastest route to the spam folder ever built — the same engagement-ratio math I walked through in the sunset flow post, now executed at agent speed. The camera shooting more footage was never the same thing as more of it being worth watching.
The team just became optional
This is the shift I actually want to talk about, because it’s bigger than one feature.
For as long as I’ve done this work, the honest answer to “should we hire an agency or a freelancer?” depended on execution capacity. An agency gave you five pairs of hands. A solo operator gave you one, and past a certain volume of campaigns, flows, and channels, one pair of hands was the bottleneck. Brands bought headcount because the work was headcount-shaped.
Composer — and the wave of agentic tooling it belongs to — changes the shape of the work. When the execution layer lives in the platform, the five-seat crew stops being the unit you hire. What you hire is the judgment: someone who knows what to brief, recognises which of the agent’s drafts are good takes and which need re-shooting, and owns the outcome when the numbers come in. One experienced person directing agents now covers the scope that justified a retainer team.
For CRM marketers and fractional CMOs, I think this is the strongest case for working independently that has ever existed. The expertise that used to be trapped inside a team structure — because you needed the team to express it — can now operate solo at team scale. The economics flip too: a brand paying for judgment plus software beats a brand paying for judgment plus four salaries of assembly work.
I’m aware I’m a fractional retention lead telling you the future favours fractional retention leads, so discount accordingly. But notice the launch design agrees with me: Klaviyo built a mandatory human approval step into the product. They’re not selling a replacement for the person who decides. They’re selling that person a crew.
The take
Use Composer. Refusing agentic tooling in 2026 is refusing the better camera. But treat every draft it hands you the way a director treats a first take from a talented crew that has never read the script: technically competent, statistically informed, and completely ignorant of what makes your account yours. Before you let it run, fix the things it will otherwise amplify: the data hygiene, the segment discipline, the authentication and deliverability posture. An agent multiplies whatever it’s pointed at, including the problems.
And if you’re the one-person team reading this: the crew requirement just dropped. What’s scarce now is the same thing that was always scarce. Knowing what’s worth shooting.