The Last 90 Days at a Company

Notice is a commit point that is hard to roll back, so the diagnosis has to come earlier. Alexander breaks dissatisfaction into tasks, influence, role conditions, environment, growth and meaning: after a product or business transformation they shift at once, and an unexamined discomfort keeps buzzing in your ear until you burn out and simply announce you are leaving. One bad week, month or project proves nothing — what counts is a red flag that repeats, its direction, and whether it blocks your goals. The 90 days, he stresses, are an illustration rather than a norm. The instrument is a journal not of irritants but of energy: what you did, with whom, in what context, how engaged you were, how you felt after. What drains you leaves your scope and what charges you is added in — a technique that works only up to a point.

The next step is to separate the negotiable from the constraints. Role tasks, interaction rhythms and authority can be renegotiated; a country's economy, company strategy — even for someone near the top team — and the nature of the business cannot: call yourself hugely technological all you like, but if the business is finance, it is obvious where the focus sits. Work is reassembled out of the set and boundaries of tasks, working relationships and meaning; the basic needs, which he hedges as self-determination theory while admitting he forgets the name, are autonomy, competence and relatedness — a sense of a shoulder beside you. His own rebuild came late last year: from CTO over roughly a thousand engineers and five large blocks into a top-grade individual contributor role — the hardest engineering and technology direction, frontier problems and models, leverage across the whole company. At that scale the work is no longer possible in Russia, hence London: the States he is not ready for, while Britain keeps relatives close and hosts offices of every company he cares about.

A list of job titles explains poorly how a person ended up where they are, so it is worth drawing a career line: turning points, the scale worked at, recurring themes, what was learned, what it cost, what contribution was delivered. The caveat: a coherent line emerges for someone who built a career deliberately rather than hopping between places — though the retrospective helps anyone. Alexander did a similar exercise for the UK Global Talent visa, which asks for a statement of why you are coming to London, what you will do and why you fit: an external body checks that the ambition is evidenced rather than fantasy. The target role should be described through accountability, decisions and outcome instead of a title, and mapped onto your development plan: the familiar question of what you can do for the company is worth inverting — what can the company do to get you where you want to be in two or three years.

The first audience question: combine the engineering and management tracks, or pick one? For Alexander it came naturally: he joined a functionally organised company after years as a developer and team lead, owned outcomes inside a product, and led by example — told that routing traffic, spinning up machines or creating test databases was technically impossible, he opened a console and made it possible. He also helped defend an internal track for high-grade individual contributors, joking that he must have known it would come in handy. Combining is very hard and demands a lot of study; whether everyone needs it he does not know, but for him technology is the leverage without which he cannot work. The second question is about managers without a technical background, and lately he has grown disappointed in them: they fit large bureaucratic organisations that maintain processes instead of delivering results, and they talk about AI reshaping development when the most they have done is ask ChatGPT something. Asked whether they have tried anything themselves, they usually have not.

Before the decision point every move should be reversible: a different task set, a new project, temporary work on a neighbouring product, a step from product into platform, a rotation into another function. Asking for a title gets you nowhere — what works is informational interviews with domain owners and an exchange of artefacts: they show how things run on their side, you show your results, successful and not. At T-Bank Alexander helped many such moves because he knew more or less everyone in engineering. A rotation suits both sides: the person skips a stressful adaptation, and the company judges by years of work rather than the weak signal of three interviews passed. The fragile part is the mutual contract: friends of his agreed on a significant change, then found after moving that authority and resources were absent and the evaluation criteria unclear. The interview runs both ways: ask where the role came from — at one startup he did not ask and regretted it — what mandate it carries, whether the manager backs it. The market calibrates: after the move he would happily see Anthropic or OpenAI from inside.

A handover is needed in all three outcomes: even a rebuilt role gives away part of its accountability. Moving into the IC role, he deliberately narrowed his scope so the change stayed local. The map covers decisions — approvals kept arriving at his desk until the last day — people and who they report to next, systems, rhythms, risks and connections; closed items are dots, yellow means in progress, red a risk with no owner. Connections are the murkiest layer of a handover: after years inside a company a person becomes a router of information who knows who should meet whom for what. The order of telling people matters — agree timing and wording with your manager first, then key colleagues and direct reports, then the team, publicly and without drama. After notice the owner becomes an adviser: explaining the history of decisions, flagging risks and showing the solution space, but not taking large commitments or deciding for the successor. Decision rights go first, with real decisions made before the final day, then context, rhythms, introductions and access; unimportant initiatives are better closed. The first 90 days elsewhere come in a separate talk: the destination is still open, from his own startup to one of the big labs.