Skip to content
All essaysA stone tower with a spiral stair seen through narrow openings and a brass telescope on the summit

Staff engineering

How to become a staff engineer: an evidence roadmap

You do not become a staff engineer by collecting difficult tickets. A nine-step plan for earning the scope before asking for the title: choose a problem that crosses teams, agree a baseline, write the decision record, secure one commitment, design adoption, and build a promotion packet with claims you can defend.

By Mike Salari. Staff mobile engineer · 500+ technical interviews from the hiring side.

12 min read

You do not become a Staff engineer by collecting difficult tickets until somebody changes your title. You get there by owning a problem that no single team can solve, changing the decision process around it, and showing that the result survives without your constant involvement. A promotion committee can assess that evidence. It cannot assess ambition.

I am a Staff Mobile Engineer and Tech Lead with 15 years in mobile, and I have run 500+ technical interviews on the hiring side.

Public career frameworks support a distinction worth making. GitLab says promotions document work already demonstrated, and its Staff guidance describes a shift toward guiding and aligning other engineers. Dropbox says its framework is not a checklist. I would not use either company's ladder as a promise about your employer. I would use them to ask better questions about scope, impact and opportunity.

This is a practical plan for producing evidence, not a guaranteed promotion in 90 days. If your organisation cannot offer cross-team scope, the plan should expose that constraint early rather than encouraging you to invent achievements.

If you are not sure yet what separates the two levels, start with staff engineer vs senior engineer. This page is the plan for producing the evidence.

The evidence packet I would actually want to read

Imagine a mobile reliability initiative. A weak promotion packet says: "Led cross-functional improvements to the mobile platform and increased team productivity." That sentence contains no baseline, accountable decision, adoption evidence or remaining risk. I cannot tell whether the author led a change or attended its meetings.

A useful packet has five links. The first is the problem statement, including who experienced the failure and how it was measured. The second is a decision record that lists one rejected alternative and why. The third is an explicit commitment from another owner, such as an agreed API contract. The fourth is evidence that at least one team adopted the change. The fifth is a post-change measure or operational runbook showing what happened after the author stepped back.

If one of those links is missing, do not camouflage the gap with promotion language. The missing artefact tells you what to do next. If the programme has not launched, you can claim a well-tested decision and pilot, not a durable outcome. If there is no accepted baseline, you can describe the implementation, not the impact. If a different team did the adoption work, credit it.

There is a hard question behind every Staff claim: why did this require somebody to operate beyond a single team's normal ownership?

Here is the kind of decision where I listen for this. Mobile and web teams fight over one API response shape, and someone proposes a backend for frontend. Who should own it? My own bias is to push ownership towards the mobile team as far as its operational maturity allows. But I have watched mobile teams sink under an edge service they were not ready to run, and in those organisations platform ownership was right until the team grew into it. There is no formula. There is the question, who needs the contract to move fastest, and the constraint, who can run the service without setting it on fire. A Staff answer does not pick a side. It names the axes and argues from both.

My suggestion is to ask your manager to review a packet before you ask for promotion. Have them point to the exact evidence the company needs and the opportunity you still lack. You are not asking for praise. You are testing whether your current work creates a credible path.

Step 1: choose the problem, not the title

The first step is to find a problem that spans more than your current team's delivery queue. That does not mean volunteering for the largest initiative. It means choosing work where the bottleneck sits at an interface between owners. Mobile examples include old-client compatibility, unbounded app startup work, fragmented observability and inconsistent offline behaviour.

Ask what failure persists even when every local team completes its assigned tasks. If nobody owns the boundary, you may have found meaningful Staff scope. But do not seize it without permission. Agree with your manager and the affected owners on what decision you may make and what outcome you may be held accountable for.

The deliverable is a one-page problem statement: affected users, baseline evidence, accountable partners, a desired result and a stopping condition. An interviewer can score a problem statement more readily than an ambition to "drive architecture".

Step 2: establish a baseline that others accept

Collect evidence before proposing a solution. If the problem is startup performance, define the measured start path, device classes, release versions and the proportion of users affected. If the problem is release safety, identify the failure classes, time to detect and recovery constraints. Do not promise a universal number that cannot be measured reliably.

The important negotiation is about evidence. Engineers, product owners and operators often disagree because they are measuring different things. Agree on which signals matter before implementation. The team should be able to tell if a change failed even if its author loves the design.

Write your baseline in the decision record and preserve the query or instrumentation method. In a promotion discussion, that baseline supports attribution. In an interview, it lets you answer the question, "How did you know this problem deserved attention?"

Step 3: write the first decision record

A useful decision record has a problem, constraints, viable options, selected option, risks, migration path, owner and revisit trigger. It does not need pages of ceremony. It needs enough precision for a team that did not attend the meeting.

For example, a mobile backend contract may need a new response field. One option is an immediate breaking change. Another is a compatibility window with additive fields and explicit deprecation. The second option costs temporary dual support. Its value depends on how many older app versions remain active and how quickly users upgrade.

Ask a peer who disagrees with your proposal to review the record. Include the objection accurately. Staff influence becomes more credible when the decision survives scrutiny without relying on your presence.

Step 4: secure one cross-team commitment

Cross-team work is not the same as copying three teams on a design document. Find the minimum commitment required from another owner. It might be a server-side idempotency contract, a common error code, a metric definition or a release gate.

Record the exact commitment, owner and date. If the owner cannot accept it, narrow the scope or propose a different sequence. Do not claim alignment because nobody objected in a meeting. Ask the decision maker to state what they are agreeing to.

In interviews I would probe what happened when the other team had a higher-priority project. A strong response shows negotiation under constraints, not a fantasy where everybody had spare capacity.

Step 5: prove the riskiest assumption cheaply

A Staff-level architectural intervention often begins with a small, disposable experiment. Build the smallest slice that tests the assumption likely to sink the proposal. For an offline queue, prove that a client crash after local commit cannot lose a pending operation. For a performance plan, reproduce the slow path on a low-end device before reorganising modules.

State the exit criterion before the experiment. If the data does not support your proposal, change course. The experiment must retire uncertainty, not create sunk cost.

This is an area where coding remains central to Staff work. A short proof can be more influential than a large presentation because it converts disagreement into evidence. The important question is whether other teams can use the result.

Step 6: design adoption and a safe exit

A technically correct solution fails organisationally when nobody can adopt it. Identify who must change client code, server contracts, testing, on-call procedures and release processes. Add version compatibility, training and rollback constraints to the implementation plan.

Treat old mobile app versions as active consumers until evidence says otherwise. App-store rollout mechanisms can limit new exposure, but they cannot instantly erase an installed version or undo a destructive database change. That constraint belongs in the migration plan, not in a footnote after launch.

A useful adoption plan includes a first adopter, migration checks, an owner for failed upgrades, a gate for widening exposure and a trigger for removing the old path. The cost of dual support is real; the plan must show who is paying it.

Step 7: make the result operable by somebody else

If your mechanism requires you to attend every review, it is not yet a multiplier. Document the rules another engineer needs to apply, create a representative example and make operational alerts understandable to the on-call rotation. Ask a colleague to perform the next change without your assistance.

This is not a demand to stop mentoring. It is a test of whether mentoring has produced independence. The evidence might be a new owner confidently resolving a known failure class, a documented release check or a decision record reused by another team.

Avoid invented productivity figures. If the record is qualitative, say so. The honest result may be that adoption succeeded in one team while a second team still needs help. Staff-level judgment includes knowing which part is not yet complete.

Step 8: build a promotion packet with bounded claims

Your packet should make the reviewer's work easy. Use a short narrative for each initiative: initial condition, decision, alternatives, people affected, measurable outcome, remaining risks and what changed after you stepped back. Attach links to the decision record, code, test, incident review or operational dashboard that supports the claim.

Separate your role from the team's role. "I designed the conflict policy and drove agreement between mobile and API owners" is sharper than "I transformed the data platform" if the second sentence overstates your remit.

Ask your manager and one stakeholder outside your team to challenge your attribution. Where evidence depends on testimony, request precise observations about decisions and changed behaviour rather than broad praise. Promotion standards vary between companies, so translate the packet into your employer's actual rubric.

The fields of each entry, and a worked example, are in staff engineer promotion packet evidence.

Step 9: practise the staff interview under changed constraints

Use your packet to rehearse two narratives. The first is a ninety-second overview of a decision and its outcome. The second is a deep dive where the interviewer can change the deadline, remove an ally, impose an incompatible old-client requirement or ask about an incident after rollout.

A weak answer treats the original design as sacred. A stronger answer identifies which invariant must remain and which cost can move. If the deadline becomes shorter, perhaps you narrow exposure rather than deleting compatibility checks. If a team refuses a dependency, perhaps you stage the solution instead of quietly assuming their work will appear.

What to do in the next 30, 60 and 90 days

In the first thirty days, identify one cross-boundary problem and collect a defensible baseline. Write down why local fixes are insufficient. Speak to the affected owners before designing a solution. Your output is a short problem statement and an agreed decision process. If you cannot obtain a real owner or a clear mandate, that is the result of the first phase. Do not hide it.

In the next thirty days, test the riskiest assumption and secure the smallest cross-team agreement needed to make progress. Record what you will not build. Publish a decision record whose assumptions can be checked by another engineer. The success condition is not that every stakeholder likes your plan. It is that a real decision can be executed and revised responsibly.

In the final thirty days, let another engineer use the mechanism, run at least one failure drill and gather early outcome evidence. If you are still the only person able to operate the system, your next task is to remove that dependency. If the initiative has not launched, do not claim the result; present the decision and the evidence you do have.

Ninety days is a review cadence, not a claim that Staff scope appears in one quarter. Some programmes require more time and several teams. The useful habit is to accumulate evidence without inflating it.

If you have just started in a staff role, the staff engineer 30/60/90 day plan lays out the first quarter week by week.

Ask your sponsor to reject one claim

Before the formal review, invite a manager or Staff peer to mark the least defensible sentence in your packet. Request a specific reason, such as missing attribution, no external adoption, a baseline measured after the work began, or an outcome that still depends on you. Replace that sentence with the narrowest true claim and plan the work needed to strengthen it. A smaller defensible claim is better than a large one that collapses under a follow-up.

The interview questions your packet should survive

"How did you decide the problem was worth solving?" requires an accepted baseline and rejected alternatives. "Who disagreed with you?" requires a real point of tension, not a staged story about an irrational colleague. "What would you do differently?" requires a limit you can name without blaming other teams. "How do you know the change lasted?" requires an owner, a metric or repeatable behaviour after handover.

A candidate can answer each of these in a minute if the evidence exists. Without evidence, another page of impressive adjectives will not help.

The Staff path is also not synonymous with management. You may coordinate many people without becoming their manager, and you may spend significant time coding. Your responsibility is to improve the decisions and systems that other people depend on. That can involve a small technical fix, a new operational contract or the decision not to migrate at all.

A final test before asking for promotion

Give your packet to somebody who was not present during the initiative. Ask them to identify what you decided, what the organisation paid for that decision, what happened after launch and what evidence is missing. If they cannot answer from your documents, the packet is not ready. That does not mean your work lacks value. It means the claim is not yet easy to assess.

Do not copy a Staff title into your CV before earning it. Describe responsibilities accurately, and show cross-team outcomes with verifiable evidence. A hiring panel can reward an honest Senior engineer with Staff-level scope. It cannot reliably score invented authority.

Questions engineers ask about how to become a staff engineer

How do you become a staff engineer?

By owning a problem that no single team can solve, changing the decision process around it, and showing that the result survives without your constant involvement. A promotion committee can assess that evidence. It cannot assess ambition.

Can I become a staff engineer in 90 days?

No plan can promise that. Ninety days is a review cadence, not a claim that staff scope appears in one quarter. Some programmes require more time and several teams. The useful habit is to accumulate evidence without inflating it.

Is staff engineer a management role?

No. The staff path is not synonymous with management. You may coordinate many people without becoming their manager, and you may spend significant time coding. Your responsibility is to improve the decisions and systems that other people depend on.

Altitude is my guide for the staff side of this: systems judgement and the written artefacts that carry it into rooms you are not in.

Share this essay