Klaviyo made agent fluency a condition of employment for 2,300 people

Klaviyo co-founder and co-CEO Andrew Bialecki told SaaStr AI that the company gave all 2,300 employees a deadline: be at "L3" agent autonomy by the end of June ^[1][2]. The interesting part is the sequencing. Agent adoption was imposed as an internal operating requirement first, and the product work followed from it.

The ladder is borrowed from levels of driving autonomy. L1 is using AI to search. L2 is spinning up a session and running an agent. L3 is constantly running multiple sessions or a team of agents, decomposing a problem, and validating the output ^[3]. It applies to every function, not just engineering: product managers, designers, sales and marketing all had the same deadline, and according to the account everyone commits code, from Bialecki down to summer interns ^[4]. The average PM who wrote wireframes and specs before AI now has to operate at L3, which is a job redefinition applied to a whole org at once rather than a tooling rollout ^[5].

The base being spent here is not small. Klaviyo reported $370.6M in Q2 '26, up 26%, with more than 205,000 customers, and raised full-year guidance to $1.526B to $1.534B ^[6]. That is a midpoint of about $1.530B ^[7], or roughly $665,000 of guided revenue per employee across 2,300 people ^[8]. Bialecki also holds an incumbent position worth protecting: roughly 80% share in the Shopify ecosystem, built on the simple move of reporting campaign revenue instead of opens, and one of the few IPOs in the 2023 cohort ^[9].

What the mandate buys shows up in the build system. Klaviyo's internal "Dark Factory" takes a prompt, acts as the PM, writes specs, decomposes the problem into engineering subsystems, writes contractual API interfaces between them, then runs subagents against each piece, building through the weekend and interrupting when requirements are ambiguous ^[10]. Klaviyo started it last fall because its early agent code was prompts stacked one on top of another into an unmaintainable mess ^[11]. The first working prototype of Composer, its marketing agent, was built over a single weekend by other agents ^[12]. Composer reached more than 95,000 users in its first month, about a quarter of them returning weekly, with credit consumption growing 30% week over week ^[13]. Sustained for four weeks, that compounding is roughly 2.9x a month ^[14], which is the kind of curve that either breaks or decays.

The product consequences are the reason the internal mandate is not theatre. Klaviyo treats the base model as a general athlete and the harness as the coaching: Composer is fed live signal on how consumers across the Klaviyo network respond, and a "coach" agent scores every proposal on predicted engagement and revenue before it ships ^[15]. Agents arrive as power users on day one and sit to the right of the best human users, so onboarding matters less and what the agent asks you to build next matters more ^[16]. Anything a human used to log into is now infrastructure and needs APIs, and Klaviyo has a dedicated engineer whose only mission is a path to sign up, configure and pay without touching the UI ^[17]. A company whose own PMs still draw wireframes cannot credibly ship that.

Bialecki says there was little pushback, because he pitched it as career insurance rather than company loyalty: few people will reach L3 in the next year or two, and those who do will do well whether they stay or not ^[18].

Watch three things. Whether Composer's roughly 25% weekly return rate holds as the 30% weekly consumption growth decays ^[13]. Whether the headless signup, configure and pay path actually ships ^[17]. And how L3 is measured, since the account of the talk does not say how attainment is verified or what happens to people who do not get there ^[19].