Your Lead Engineer Just Resigned Mid-Release: The Next 72 Hours

Last updated: November 2, 2026

By Mike Carter, Managing Director, KORE1

When a lead engineer resigns mid-release, spend the first 72 hours naming an acting release owner, deciding whether to ship, shrink, or slip, inventorying what only that person holds, and opening a contract search for the acting lead’s hands-on work. Once an offer is accepted, KORE1 contractors start a median of 3 days later, and 86% are working within a week (n=1,997).

The email lands at 4:40 on a Tuesday. Code freeze is nine days out. The person resigning wrote your deploy pipeline, holds the only admin login on two vendor accounts, and is who everybody asks before merging anything scary. Two weeks’ notice. Very polite. There’s even a line about wanting to make this as smooth as possible.

I made that Tuesday up. You probably didn’t need me to.

So here’s where I sit, before you take advice from me. I’m Managing Director at KORE1, a staffing firm, and we get paid when you bring in contract staffing through our IT staffing services desk. I’m also not an engineer, and you know your system better than I ever will. What I can tell you is the staffing half of this, including the part where staffing is the wrong place to start.

Hour zero is not a hiring problem. Hour 48 might be.

What Should You Do First When a Lead Engineer Resigns Mid-Release?

Name one person as release owner the same day, before anyone writes a job description or books an exit interview.

Key-person release risk is the chance that a release can’t ship, or can’t be rolled back, because the knowledge or access it depends on sits with one person. A resignation doesn’t create that risk. It puts a date on it, usually about two weeks out, and that date has no reason to line up with your release.

Google’s Site Reliability Engineering book splits an incident into separate roles: a commander who holds the overall state, an operations lead, and someone handling communication. Nothing is down here, I know. Copy it anyway, because two directors each quietly running the response will hand your team two answers by Thursday. The same chapter is blunt about handoff, too. The outgoing commander says “You’re now the incident commander, okay?” and doesn’t leave until somebody acknowledges it. Have your resigning lead do the human version of that. Out loud. In front of the team, with the acting owner’s name in the sentence, so that nobody on Slack on Wednesday morning has to wonder who they’re supposed to ask about the release branch.

The whole 72 hours, then. It fits on a screen.

Ship, Shrink, or Slip: Make the Release Call by Hour 24

Pick one within a day. Why so fast? Every hour without a call gets burned by engineers guessing which version of the release they’re building, and they will guess differently. There are three options. People like to propose a fourth, where everybody just works harder and nothing changes, but that one’s really the first option with worse information and a tired team.

Me? I lean shrink. It keeps a date alive, and there’s usually something in the scope a customer wouldn’t miss for six weeks. Ask product. They’ll know, and they’ll hate it. Fine.

One rule beats the table. Nobody left who can roll it back? Don’t ship it.

I’d hold to that even against the departing lead, whose sincere offer to “just get it over the line” runs out on a known Friday, and who won’t be around the Tuesday after, when the migration you shipped starts doing something odd to the invoices table.

What’s in Their Head, and What’s in Their Name?

Two lists. Everyone builds the first one. The second one is what breaks, quietly, three weeks later.

Start the head list in version control. Skip the meeting. Run git shortlog -sn --since="12 months ago" HEAD -- path/to/service against each service in the release and you’ll see who actually touched what. Check your GitHub CODEOWNERS file too. Code owners get requested on pull requests automatically, so that name is about to become a review queue nobody reads. One person behind most of the commits on something in this release? Book a capture session.

The name list is harder. It’s credentials and accounts that got set up under one person because, on the day, that was fastest.

• AWS first. Pull an IAM credential report, and know its limit, since it covers only the first two access keys per user and skips service-specific credentials entirely.
• CI that runs on a personal access token, so the pipeline dies when the account does. GitHub’s deploy key documentation lays out the alternatives, including a machine user. Note that a deploy key stays active after the person who made it is removed, so it won’t break and it won’t go away either.
• Domain registrar. TLS certificate renewal emails going to one inbox. Who’s the Account Holder on your Apple Developer account?
• Terraform state, and whether anyone else has ever applied it.
• PagerDuty schedules and escalation policies with that name at the bottom.
• The vendor rep at Datadog or Snowflake who only ever emails your lead back. Nobody writes this one down, and it’s the one you’ll feel at renewal.

Don’t pull access on day one for someone working out a friendly notice, though. You need them logged in. Inventory now, then remove everything on the last day in one sitting. NIST SP 800-53 Rev. 5 has a control for exactly this, PS-4, which has you disable system access within a time period you define and keep access to whatever systems the person controlled. That second half gets missed. Headed to a direct competitor? Different math. Call counsel before you read another staffing blog.

How Do You Capture Knowledge From a Departing Engineer in Two Weeks?

Make them watch somebody else do the work, and let them speak only when that person gets stuck.

Recordings of the lead walking through their own system feel productive. They aren’t, mostly. You get a confident person skipping whatever’s obvious to them, which is exactly what you needed. Flip it. Have the acting owner run the deploy, and the rollback, while the departing lead sits on their hands, and write down each time they lean in and say “oh, you have to also.” Each of those sentences is a runbook line. Two or three of those sessions will beat a 40-page document written in a hurry by someone whose mind is already at the next job, which nobody opens until something’s on fire anyway.

DORA’s research on documentation quality finds a clear link between documentation and organizational performance, and it says the part everyone skips, which is that documents have to be actively maintained. So forget complete. Aim for tested by someone who isn’t leaving.

Questions I’d want on the list, from people who know these systems better than I do:

• What do you check by hand before a release that no alert checks?
• Which thing looks broken but is fine?
• If this pages at 2 a.m., who would you call, and is that person you?
• What did you promise a customer, a vendor, or a director that isn’t written in any ticket?

Some of this already has a home on our site. The contractor handoff template was built for contract endings, but its runbook, ownership map, and open risks register work fine for an employee too, and Tom Kenaley’s piece on proving a knowledge transfer with a test makes the same argument at more length.

Is it survivable if you get this wrong? Usually. A study of 1,932 popular GitHub projects by Avelino and colleagues found 315 of them, 16%, were abandoned by their core developers, and 128 of those, 41%, survived because new core developers took over. Open source isn’t your payroll. Nobody there had a code freeze. But look at what the survivors had. Somebody new who could get in.

Who Do You Actually Bring In? Backfill the Acting Lead, Not the Lead

Bring in a contractor for the acting lead’s hands-on queue, not for the lead’s chair.

Fred Brooks wrote in The Mythical Man-Month in 1975 that adding manpower to a late software project makes it later. Still true. And I sell manpower. A stranger dropped into a release nine days from freeze will consume your best remaining engineer’s time just getting oriented, and that’s the exact person you can’t spare, the one who picked up a second job on Tuesday.

So don’t aim the stranger at the release. Your acting owner’s old tickets, their reviews, and the feature they were building are all sitting unowned now, and those are ordinary, bounded pieces of work with tests around them that someone competent can pick up without knowing why the billing service was built that way in 2021. Our backfill staffing page covers the general version of holding a seat. Here, the seat you hold isn’t the one that emptied. When the crisis is a failing system rather than an empty chair, Tom Kenaley makes the same call for a production incident that outlasts your team.

Which contractor? Depends what the lead was really doing all day. That’s rarely what the title says.

Source: KORE1 Contract Rate Index 2026-Q4, effective October 16, 2026. Ranges run p25 to p75. Three months is 520 hours at the median. Modeled cells use BLS OEWS May 2025 wages, not KORE1 placements. Next refresh January 2027.

Mind that last row. A contractor whose end date everybody can see shouldn’t be deciding things your team lives with for five years.

Four weeks of a $120 engineer is $19,200. Put that against what your slip costs. Sometimes the slip is cheaper. So slip.

How Fast Can a Contract Engineer Really Start?

Faster than a permanent hire, and not by hour 72.

Across 1,997 KORE1 contract placements from 2019 to 2026, the median start came 3 days after an accepted offer. The full distribution shows 37% starting the same day and 86% inside a week. Read that with care, because the sample covers every role family we place, it’s 79% California, and the clock only starts once someone says yes. Our 17-day average time-to-fill for IT roles is a separate clock measuring a separate thing, so a real start is the search plus the start, those two figures don’t come from the same placements, and neither number is a promise. We don’t guarantee dates. Nobody honest does.

So your hour-72 target isn’t a person at a desk. It’s a brief good enough that a recruiter can move today.

• Stack, precisely. “Python 3.12, FastAPI, Postgres on RDS, deployed with GitHub Actions to EKS” gets you a shortlist. “Backend” gets you resumes.
• The freeze date, and what this person will and won’t touch before it.
• Who reviews their first pull request. By name.
• Remote or onsite, and whether a security review stands between yes and day one. Our SOC 2 provisioning guide covers what access looks like if an auditor will ask.

Open the laptop and access tickets the same day you extend the offer. Then hand them the 30-day ramp plan, which puts a first merged pull request in week one.

Cost-wise, the contract staffing cost calculator will run your metro. Duration-wise, our median contract assignment lasts 116 days (n=1,887), which roughly covers a permanent search with room left over for overlap. If you might keep the person, structure it as contract-to-hire, where conversion carries no fee after 1,040 hours and 37.5% of our contract-to-hire placements did convert (207 of 552). Otherwise run a direct hire search alongside it for the real replacement. Different track, different clock.

When shouldn’t you call us? If the release shrinks cleanly, the acting owner’s queue can wait six weeks, and you’ve got a strong internal candidate, then you don’t need a contractor, you don’t need us, and the money is better spent on whoever just took on two jobs. Promote the person and take a breath. I’d rather lose that invoice than have you pay for a seat nobody needed.

The Conversation You Have With Everyone Who Stayed

Tell the team yourself, the first day, with the acting owner’s name and the date of the release call. They already know. Engineers talk.

Then the quieter part. Your two or three strongest remaining people are doing math about their own workload, and maybe about what the lead knows about the company that they don’t. Have those conversations one at a time, inside the first 24 hours, and say out loud what’s coming off their plates, because a person who can see their own workload going down is a lot less likely to start returning recruiter calls this week. Kris Drouet’s post on backfilling engineers after an acquisition covers departures that arrive in stages, and the same worry applies here. One resignation is a bad week. Two in the same release is a different problem, and that second one is the one you can still do something about.

What CTOs Ask in the First Week

Should I counteroffer, or does that just delay it?

Only counter if the reason they’re leaving is something you can actually fix, and the fix survives the release. Money solves a money problem, and plenty of resignations aren’t about money. A stay bonus paid on a date after the release is often the more honest deal, because it buys you the weeks you need, it’s easy to write down, and nobody has to pretend the relationship has been repaired. Keep running all the steps above either way.

They gave no notice and walked out today. Now what?

Cut their access first, today, then run the same plan compressed. You’ve lost the capture sessions, so your knowledge list comes entirely from version control, the ticket history, and the people who sat beside them and will, if you ask nicely, remember more than they think. Slip becomes the likely call for anything with a migration in it.

Can the person who resigned stay on as a paid consultant?

Sometimes, as a small, written, time-boxed agreement for questions, though their new employer’s paperwork may limit it. Have counsel look at it. A former employee doing the same work on a 1099 is a worker classification question, and the IRS weighs behavioral control, financial control, and the type of relationship. Ten hours of answering Slack questions looks very different from staying on the on-call rotation.

Realistically, can a contractor just become the lead?

Not usually, and not for this release. A senior contractor can carry delivery, mentor people, and run a workstream, but the authority to make calls that outlast their contract should sit with an employee who’ll still be around to answer for them. A contract-to-hire conversion later on is a different conversation.

$62,400 for three months. Is that actually cheaper than waiting?

Depends what the wait costs, and only you know that number. That figure is 520 hours at the $120 median we’ve seen for senior software engineers (n=7, primarily Southern California), and it buys capacity, not a guarantee the date holds. Write down what one month’s slip costs in revenue and contract terms. Then compare. Can’t name a cost? You may not need the contractor.

Start With the Date

Resignation letter still open in another tab? Close it. Name the owner. If the math then says a seat needs covering, send our team your freeze date and the stack, and we’ll say plainly whether a contractor can help in time, or whether you’d do better to shrink the release and call us once it’s out. Later, when you’ve slept, our dependency risk scorecard can help find the next single point of failure. Preferably before it finds a new job.