Saying no, the most leveraged act
Every yes is an allocation. Saying no well is what separates leaders from order-takers.
For an engineering leader, saying no means turning down or reshaping a request for the team's time so that the work the team has already committed to still gets done. The standard advice, and the one I use, is a tradeoff script: a team has a fixed amount of time, so every yes pushes some other work out, and when a request arrives you name the work it would push out and offer something smaller or later in its place. Michael Porter made the same argument about companies in 1996 when he wrote that the essence of strategy is choosing what not to do, and Greg McKeown's Essentialism later applied it to one person's calendar. The script works because the person asking can see the tradeoff and help make the decision, which is a better conversation for both of you than a flat refusal.
You use the script when you refuse, but nobody uses one for the requests a leader accepts without much thought, and the time for those requests still has to come out of something.
Take an engineering manager at a payments startup whose quarter is built around rewriting the payments service. In the first month, a large customer asks for a weekly custom data export, and the customer success team brings two smaller requests of about two days each. The manager accepts all three, because none of them looks big enough to argue about, and the sprint plan stays as it was. Over the next six weeks the rewrite's date moves by a week at a time, and the two engineers on it stop writing tests for the new code so that each slip stays small. Nobody refused the export or the two smaller requests, but the engineers did refuse the tests, and they made that call on their own, without the context the manager had when the requests came in.
A request outside the plan needs the same amount of time whether or not anyone argues about it, so if the leader does not decide where that time comes from, the team decides, and the team takes it from work that nobody checks:
- tests for new code
- documentation
- a code review that gets a quick approval
- a date that moves a week at a time
In the payments example, the customer and the customer success team got their yes on the day they asked, and the bill went to two engineers who were never part of that conversation. So when the plan looks the same on paper after several accepted requests, those four are the places to look. The leader is choosing between one no, said early and in the open by the person with the most context, and several smaller nos said later by the team where nobody sees them.
Work out the cost before you answer
A yes is easy to give because the person asking is right in front of you and the cost arrives weeks later, on someone else's desk. So before you respond, answer two questions: what current work would this push out, and who will own the result after it ships. The first answer gives you a tradeoff you can show the person asking, and the second matters more than it looks, because a feature or an integration needs care for as long as it exists and some person on the team inherits that care. Plenty of requests pass both questions, so you can do this in your head for anything small and save the written version for the large ones.
A founder sends a screenshot of a competitor's new feature at ten on a Sunday night, which is when founders find competitor features, with one line asking whether the team can build it. A manager who once would have replied yes before Monday's standup instead replies on Monday afternoon with the cost: the feature would push the checkout redesign out by three weeks, and one of the two mobile engineers would own it afterwards. The founder reads the reply, decides the feature can wait a quarter, and it goes into next quarter's planning.
Say no early, to the person who can decide
A request that would push out committed work needs an answer in the same week, given to the person who has the authority to accept that tradeoff, and that person is not always the one asking: a head of sales can ask for a custom integration but cannot decide that the onboarding work moves, so the tradeoff goes to whoever can decide, written down in a paragraph. "Yes, if" is a fair answer, as long as the work named after the "if" leaves the plan. When a leader says yes on a condition and then lets the team absorb both pieces of work, the requester learns that the leader's limits are an opening position, and the next request comes sooner.
Sales is close to signing a big retail chain, and the client wants a custom invoice format and an integration with their accounting system within three weeks. The engineering manager works out the cost: the onboarding revamp planned for the quarter would slip by a month, and the integration would need an owner for as long as the contract runs. That same week, the manager sends one paragraph with both costs to the founder and the head of sales. The founder chooses the deal and tells the team the same day that the revamp moves by a month, with a new date, so no engineer spends that month trying to carry both projects.
If the person with authority decides against your recommendation, you have still made the cost visible, and a named person now owns the decision. After losing that argument you still have to commit to the decision without pretending you agreed, which is its own practice, and it helps to write the decision down so the same request does not come back next quarter without its cost.
Give the ownership reason
The usual reason behind a no is capacity: the team does not have the time. A requester who has watched a coding agent build a working prototype in an afternoon will fairly ask why a small feature needs a month, and "we do not have the time" stops sounding true to them. Answer with ownership instead, because the afternoon pays only for the build, and after it the feature needs:
- support when customers use it
- maintenance when the systems around it change
- someone's attention when it breaks
- room in a product that now does one more thing
Each of those lands on a person on the team for as long as the feature exists, and that person's time is the reason you give for the no.
An operations team at a logistics startup asks for a tool to update merchant records in bulk, and an engineer builds it with a coding agent in one afternoon. Six months later there are three versions of the tool, each made for a slightly different request, and nobody is named as its owner, so every change to the merchant data model now depends on someone remembering that the tools exist. Nobody asked who would own the first version, because an afternoon of work did not seem worth the question.
You can ask the same question to find the cheap yeses: an experiment the team intends to delete has almost no ownership cost, so it can get a yes with a delete date written next to it. If the date passes and the experiment is still running, name an owner for it or delete it that week.
Write down what you will not do
When the same kinds of requests reach you every week, refusing each one in person costs you time and costs the requester a small disappointment each time. A short list titled "not doing this quarter", published next to the quarter plan, lets you make those decisions once; choosing what goes on it is the work of deciding what gets starved on purpose. Pair the list with a single place where new requests go and a fixed weekly review where they get weighed together, and fewer requests arrive as interruptions. The list only helps if people read it and if the weekly review gives every request an answer, because a request that goes into the intake and never comes out leaves the requester waiting and the team guessing whether it will come back.
Match your nos to your track record
When a leader says that a request would push the release by two weeks, the people hearing it test that claim against the leader's earlier estimates. A manager whose estimates have held for a year gets believed quickly, while a manager in the first quarter of the role has no such record yet, so each refusal gets judged on its argument alone. A new manager should still refuse some requests, keeping those few refusals for the requests with the largest cost and delivering the accepted ones on the promised dates so the record starts to build. Nobody can give you a count for how many refusals a new manager can afford in a quarter, because the answer changes with the people asking and with how the first few deliveries go.
A senior engineer becomes an engineering manager, and in the first quarter the head of product sends several requests. The new manager accepts two small ones and delivers both on the promised dates, then declines a larger third request with the tradeoff written out. The head of product accepts the refusal, partly because of the argument and partly because the two deliveries were recent.
A new manager can also turn "I have not earned it yet" into a permanent reason, but after two or three quarters the record exists, and a manager who still refuses nothing has gone back to letting the team say the no. Refusing someone senior to you also means making a case they will accept, which is its own skill.
Know when to say yes against the plan
Some requests deserve a yes even when they break the plan: a regulator's notice that sets a fixed deadline for how payment data is stored leaves a fintech team no real choice, and a deal that would change the company's year can justify moving committed work. The yes is right in both cases, and you still work out the cost: name what drops out of the quarter and tell the team before planning, so the cost shows up in the plan instead of in the team's evenings. Teams that hold some capacity in reserve have less to drop when these arrive.
A manager can make the same call at a smaller scale: two weeks before its Diwali sale (the biggest shopping week of the year for many Indian consumer apps), a team that has planned a code freeze gets two requests from marketing: a new coupon type and a spin-the-wheel game. The manager says yes to the coupon, which is small and fits inside an existing checkout flow, and no to the game, which would need new screens during the freeze, and offers the game a place in the plan before the next sale.