Read and ask in week one, test what you heard in weeks two and three, set direction by month one. AI compresses about half the work.
Bottom line. Spend week one building a picture of the system you now run, and change nothing. Use an agent over the team’s documents, tickets, and decision records to compress the reading from weeks to a day. Spend the time you saved on the half that does not compress: people telling you what the documents left out. Ask every direct report the same questions so the patterns are visible. By the end of month one, name the north star if one exists and write it if one does not. Pick one early win that matters to your manager. Trust comes from what you do with the first thing someone tells you in confidence. Nothing shortens that.
Why the first month is for learning
Conant and Ashby proved in 1970 that anyone controlling a system needs a working picture of it, and that the quality of the control is limited by the quality of the picture.
A new manager has no picture. Every instinct that worked on the last team now gets applied to a system whose failure modes, history, informal authority, and unwritten rules are unknown. The first month is for building that picture. Treating it as anything else produces confident decisions from an empty one.
Michael Watkins put a number on the cost of doing this slowly. In The First 90 Days he describes the breakeven point, where a new leader produces as much value as they consume. A mid-level leader takes roughly 6.2 months to get there. A structured approach cuts that by around 40 percent.
The 6.2 months aggregates executive estimates rather than accounting, and it varies enormously by situation. The number is soft. You are underwater for months either way, and how fast you surface depends on how deliberately you learn.
Week one: read and ask, change nothing
The temptation in week one is to demonstrate value. Resist it. Anything you change in week one comes from an empty picture. The people watching will read it as performance rather than judgment, and they will be right.
Two things belong in week one.
Read the record. Design documents, postmortems, the last four quarters of planning artifacts, the ticket backlog, and whatever decision log exists. You are looking for what the team has already tried, what broke, and what got decided and never revisited.
Start the same-questions round. Meet each direct report individually and ask all of them an identical set. Watkins recommends this for an analytical reason rather than a social one. Identical questions make patterns visible. Five people independently naming the same broken thing is data. Five different complaints is a different situation entirely.
A workable set. What should we keep doing. What should we stop. What would you fix if you had my job for a week. What do you think I am going to get wrong. Who outside this team do we depend on most.
The fourth question returns more than the others. It gives people permission to warn you, and the answers tell you what the team expects from a new manager, which is usually shaped by the last one.
Weeks two and three: find out what you got wrong
By now you have a picture assembled from documents and first conversations. It is wrong in ways you cannot yet see. Weeks two and three are for finding out where.
Go find who decides. Watkins makes this point and it matches the research on organizational models: the org chart is reliable about accountability and escalation and unreliable about influence. The person whose opinion ends an architecture debate is frequently not the person the chart names. You learn this by watching a decision get made, not by reading a document.
Attend the meetings where the real work gets done. The incident review, sprint planning, the design review. Say as little as possible. You are measuring three things: how the team handles disagreement, whether bad news travels, and how long it takes someone to say “I do not know.”
Talk to the dependencies. The teams you rely on and the teams that rely on you will describe your team differently than your team describes itself. That gap will teach you more than anything else this month.
Meet your manager and agree on their definition of success. Watkins ranks this relationship first for reaching breakeven, and the failure mode is assuming alignment rather than writing it down.
Month one: name the direction and pick one win
By the end of the first month you owe the team two things.
A north star, stated in a sentence. Ask first whether one already exists. A team that has one and feels unheard will resent a new manager replacing it. If a direction exists and still describes the work, say it out loud and commit to it publicly. If none exists, or the stated one has stopped describing what the team does, write one and say outright that it is new.
The test for a usable north star is whether it settles arguments. A direction nobody can use to decide between two options is a slogan.
One early win, chosen carefully. Watkins is specific about the traps: trying to do too much, and picking a win that matters to you rather than to your manager. Choose something visible, achievable inside a few weeks, and aligned with what your manager is measured on. The win matters less than the credibility it buys, and that credibility funds the harder change in month three.
What to hand to an agent
The reading half of this has changed a great deal.
A new manager’s core problem is context deficit. The team’s history sits in design docs, tickets, incident write-ups, Slack threads, and planning artifacts across several systems. None of it is indexed together, and all of it takes weeks to read.
An agent pointed at that corpus compresses the reading substantially. Useful questions to ask it. What has this team attempted here before. What decisions were made and later reversed. Which components generate the most incidents. What commitments exist to other teams. What themes appear across the last year of retrospectives.
The retrospectives question returns the most. A year of them contains a pattern nobody has assembled. Each one was read once, by the people who attended it.
Two cautions apply, and a new manager is more exposed to both.
Documents are the data, people are the anecdotes. Bezos’s rule: when the two disagree, the anecdotes are usually right. The disagreement means something is wrong with the measurement. For a new manager the document record is incomplete in a specific direction. It captures what was decided. It omits what was felt, who objected, and what the team knows but never wrote down.
Do not arrive with conclusions. The failure mode is showing up in week one with a synthesized view of the team’s problems and presenting it. That reads as a manager who has already decided, and it closes the conversations that would have corrected your picture. Use the compression to ask better questions, not to skip the asking.
Trust takes the time it takes
Trust comes from a sequence of small tests. No tool shortens the sequence.
Someone tells you something in confidence in week two. What you do with it determines whether anyone tells you anything in week six. Someone raises a problem and you either follow up or you do not. You commit to something small and either deliver it or explain why not.
Ambady and Rosenthal’s work on thin-slicing found that judgments form within minutes and then resist revision. That is why early wins and early behavior matter out of proportion to their size.
Do the small things you said you would do, on the day you said you would do them, for the first month. That is most of trust, and it is unglamorous work.
The one-page plan
Week one: read the record, ask every report the same five questions, change nothing.
Weeks two and three: find who decides, sit silently in the team’s real meetings, talk to every dependency, and write down what your manager means by success.
Month one: state the north star or write one, and land one early win that matters to your manager.
Throughout: use an agent to compress the reading, spend the saved time on people, and treat the document record as incomplete rather than authoritative.
Reader discussion