Managing former peers, and glue work

What it is

Two problems that share a root: work and authority that exist but are not acknowledged.

Managing former peers is the transition from being one of the group to holding information and decisions the group does not have. The relationship does not survive by being left alone, because it has changed whether or not anyone says so.

What actually changed on day one:
  - you know things they do not (compensation, performance
    conversations, reorganisation plans, who is leaving)
  - your casual opinion is now a directive, whether you meant it
    that way or not
  - complaining to you is now a risk for them, so you stop
    hearing the complaints
  - your friendship has an asymmetry it did not have before

Glue work, in Tanya Reilly's framing, is the coordination, communication and maintenance work that makes a team function and that no promotion rubric rewards.

Examples: noticing the design gap between two teams, writing the
onboarding doc, running the incident retro, unblocking someone
by finding the right person, keeping the roadmap coherent,
answering the same question for the fifth time, updating the
runbook, chasing the dependency.

Properties: essential, invisible, uncredited, and it accretes on
whoever is most conscientious, which is not randomly
distributed.

What this is confused with: glue work and low-value work. It is high-value work with no attribution mechanism. The failure is not that people do it; it is that they do it instead of the work that gets them promoted, and nobody made that trade deliberately.

Also confused: managing former peers and being liked. A lead who optimises for the friendship being unchanged will avoid the conversations that make them a lead, and the team notices that faster than the lead does.

The problem it solves

The former-peer transition fails silently, in one of two directions.

DIRECTION 1: nothing is said, so nothing is renegotiated.
  Week 2  a former peer vents about another engineer, as they
          always did. The lead now cannot un-know it, and must
          either act on it or explicitly not act on it. Either
          way the friendship has been used as a channel.
  Week 6  the lead makes an offhand comment about an approach.
          The engineer treats it as a decision, because it is
          one now. The lead is surprised. The engineer is
          annoyed at being second-guessed later.
  Week 12 the lead gives the interesting migration work to
          their closest former peer, because they trust them.
          Two other people notice. Nobody says anything, and the
          lead's read is that everything is fine.

DIRECTION 2: overcorrection.
  The lead becomes distant and formal with everyone, treats
  their former closest colleague more harshly than others to
  prove impartiality, and loses the relationships that made
  them a credible lead in the first place.

The second failure is more common among conscientious people and is more damaging, because distance is read as a change in respect rather than a change in role.

And the glue work failure has a measurable shape:

A team of eight. One person, over a year:
  - wrote and maintained the onboarding doc
  - ran every incident retro
  - was the informal answer desk for the legacy system
  - noticed and fixed three cross-team design mismatches before
    they shipped
  - kept the roadmap page current

At promotion time their packet said: "highly trusted, the
person everyone goes to."

The calibration room asked for evidence of technical scope. It
did not exist, because they had spent the year doing the work
that produced no artifact.

The team could not function without them and the process could
not promote them, and both statements were true simultaneously.

This distributes unevenly along predictable lines, which is what makes it a fairness problem and not only a career one: the person who notices the gap and quietly fills it is disproportionately likely to be a woman or someone from an underrepresented group, and the promotion consequence compounds.

Mechanics

The transition conversation, done explicitly and early

Have it individually, in the first two weeks, with everyone, including the people you were not close to.

The four things to say out loud:

1. WHAT CHANGED, NAMED.
   "I now know things I can't share, and I'll say 'I can't talk
    about that' rather than deflect, so you'll know the
    difference between me not knowing and me not being able to
    say."

2. WHAT DID NOT CHANGE.
   "I still want you to tell me when I'm wrong. That is more
    useful to me now, not less, and it will get harder for you
    to do, so I'm going to keep asking."

3. THE WEIGHT PROBLEM.
   "My opinions now land heavier than I intend. If I say
    something that sounds like a decision and I meant it as a
    thought, ask me which one it was. I will not be annoyed."

4. THE ASK.
   "What would you want me to do differently now that I'm in
    this role? And is there anything about this that's
    awkward?"

The fourth question is the one that produces information, and
only if you ask it in week one, before there is a specific
incident it could be about.

The specific version for a close friend, held separately:

"Two things I want to say out loud. One, I'm going to be
 careful about fairness in a way that might feel like distance,
 and it isn't. Two, if you ever feel like I'm treating you
 harder than others to prove a point, tell me, because I might
 be and I won't notice."

Naming the overcorrection risk in advance is what makes it
correctable later, because you have given them permission to
raise it.

The first-quarter fairness discipline

Everyone is watching for favouritism, and they are watching the assignments rather than the words.

For the first quarter, over-index on visible fairness:

- ASSIGNMENT: write down who got what interesting work and
  check the distribution monthly. Not because you are biased,
  but because you cannot see your own pattern and other people
  can.
- ACCESS: if one person's 1:1 regularly runs long and others
  end on time, that is visible. Equalise it deliberately.
- INFORMATION: do not let your former close colleagues learn
  things first because you talk to them more. Announce to the
  group, then discuss individually.
- SOCIAL: continuing to have lunch with the same two people
  every day is a signal, whether you intend it or not. It is
  worth changing for a quarter and then relaxing.
- CORRECTION: the trap in the other direction. If you find
  yourself giving your friend harder feedback than the same
  behaviour would earn from someone else, that is
  overcorrection, and it is unfair in a way that is harder to
  see.

And the trap that is specific to a newly promoted technical lead:

KEEPING THE FUN WORK.

You were promoted because you were good at the technical work,
you still enjoy it, and there is a plausible story ("this is
the risky part, I should do it").

Every time you do, you take the most developmental piece of
work away from someone who needed it, and you signal that the
interesting work is not available to them.

The test: over a quarter, what fraction of the most
technically interesting work did you keep? If it is more than
roughly a tenth, that is the pattern, and it is the single
most common failure of a first-time technical lead.

Seeing glue work

You cannot distribute or credit what you cannot see, and it is invisible by construction.

The audit, once a quarter, half an hour:

For the last quarter, list the things that happened that were
not on anyone's roadmap:
  - who onboarded the new hires?
  - who ran the retros?
  - who noticed the cross-team problem before it shipped?
  - who answers the questions in the team channel?
  - who updated the runbooks after the last incident?
  - who chased the dependency?
  - who wrote the doc nobody was assigned to write?
  - who organised the offsite, the on-call swap, the rota?

Then count, per person. The distribution is almost never even,
and the lead's prior about who is doing it is frequently wrong,
because the whole property of glue work is that it is done
quietly.

The three responses, and all three are needed

1. DISTRIBUTE IT.
   Rotate the visibly non-promotable parts: retros, onboarding,
   the answer-desk role, the runbook. A named rotation removes
   it from "whoever is most conscientious", which is the actual
   allocation mechanism today.
   Some of it should also just stop: a meeting that only
   exists because someone keeps organising it is a candidate
   for deletion rather than rotation.

2. MAKE IT VISIBLE AND CREDITED.
   Glue work with an artifact is promotable; glue work without
   one is not. So convert where you can:
     answering the same question 5 times  -> the FAQ, with
                                             their name on it
     noticing the cross-team gap          -> a written design
                                             note circulated to
                                             both teams
     running retros                       -> a written pattern
                                             analysis across a
                                             quarter of
                                             incidents
   This is not cosmetic. The written version is genuinely more
   valuable, and it is also the version that survives to
   calibration.

3. PROTECT THE PROMOTION PATH.
   Tell the person explicitly: "you're doing a lot of the work
   that keeps this team running, I see it, and it will not get
   you promoted on its own. So I'm rotating half of it away
   from you, and I'm giving you [named technical scope] which
   will."
   Saying it out loud is necessary, because the person doing
   the glue work usually believes it is what a good colleague
   does, and stopping feels like letting people down.

Point 3 is the conversation people avoid because it sounds like devaluing the work. The framing that works is that the work is valuable, the reward system does not see it, and the lead's job is to manage that gap rather than to pretend it does not exist.

And in the calibration room, value it explicitly:

Not: "they're the glue of the team."   (an adjective)
But: "they built the onboarding path that took our
      time-to-first-commit from 19 days to 3, ran the retro
      programme whose action-item completion rate is 84%
      against an org average of 40%, and caught two cross-team
      design mismatches before they shipped, one of which the
      platform team's own review had missed."

Same work. The second version is scope evidence.

A worked example: a promotion, a rotation, and a friendship

An engineer promoted to lead of the team they had been on for two years, with eight reports, two of whom were close friends and one of whom had also applied for the role.

The first two weeks: individual transition conversations with all eight.

The four points, said the same way to everyone. Then the
question: "is there anything about this that's awkward?"

Answers worth recording:

The person who had also applied: "Honestly, yes. I'm fine with
  it, and I'd like to know what you think I'd need to be
  considered next time."
  -> that is a career conversation the lead could not have
     initiated without it being patronising, and it became a
     concrete plan.

A close friend: "I'm worried you'll be weird with me."
  -> which is exactly the overcorrection risk, named by the
     other person in week one, and the lead's answer ("I might
     be, and you should tell me") made it correctable.

A quieter engineer the lead had not been close to: "It'll be
  fine as long as decisions get explained. Before, things just
  happened and nobody said why."
  -> a specific process request from someone who would not have
     volunteered it unprompted, and it became the team's
     decision-log practice.

Three of eight conversations produced information the lead would not otherwise have had, and all three came from the same closing question.

The first-quarter fairness data:

The lead tracked assignment of "interesting" work (defined in
advance as: new systems, cross-team scope, anything with
external visibility) for one quarter.

Month 1, before tracking:
  of 6 such items, 4 went to the two former close colleagues.
  The lead's own read at the time was "they were the right
  people".

That distribution was not visible until it was written down,
which is the point of writing it down.

Months 2-3, corrected:
  of 9 items, 2 to those two, 7 distributed. Two went to people
  who were slower at them, which cost time and was the price.

Four of six is not a subtle bias and the lead did not see it, which is the general argument for measuring assignment rather than trusting the read.

The fun-work trap, caught by a report:

Month 4, in a 1:1, an engineer said: "you took the rate-limiter
rewrite. That was the piece I wanted."

The lead's justification had been that it touched the payment
path and was risky. It was also the most interesting thing on
the roadmap.

Count for the quarter: of 5 items the lead considered the most
technically interesting, they had kept 2.

Handed it over, with two weeks of pairing. It took longer. The
engineer's own later feedback was that it was the thing that
made them believe they had a future on the team.

Keeping two of the five most interesting items is 40 percent, which is far past the tenth this page recommends and was invisible until someone said it.

The glue-work audit, run at the end of the first quarter:

Non-roadmap work in the quarter, by person:

  Priya    onboarding doc + onboarded 2 hires, ran 5 of 6
           retros, answers ~70% of the team-channel questions,
           maintains the runbooks, organised the on-call swap
           twice
  Jo       ran 1 retro, wrote 1 runbook
  Alex     none identified
  Sam      chases the identity-team dependency weekly
  (four others: negligible)

Priya was carrying roughly 60% of the identified glue work.

Priya's technical output that quarter was the lowest on the
team, and the lead's prior read of that had been "steady but
not stretching".

"Steady but not stretching" was a misreading of someone doing 60 percent of the team's invisible work, and it would have become a calibration statement.

The three responses, applied:

DISTRIBUTE: retros became a named rotation, one per person per
  cycle. The answer-desk role became a weekly rotation (this
  also became the interrupt shield). Onboarding became a named
  buddy assignment per hire rather than defaulting to Priya.
  One recurring meeting Priya had been organising was deleted
  entirely, because nobody could say what it was for.

MAKE VISIBLE: Priya's onboarding work became a written
  onboarding path with her name on it and measured outcomes
  (time-to-first-commit 19 days -> 3). The retro work became a
  written quarterly pattern analysis, which found that 4 of 11
  incidents that quarter had the same root cause, which nobody
  had noticed one retro at a time.

PROTECT THE PATH: the explicit conversation.
  "You're doing about 60% of the work that keeps this team
   running. I see it, it's valuable, and it will not get you
   promoted on its own, which is a failure of the system rather
   than of you. So I'm rotating half of it away, and I'm giving
   you the streaming migration, which is the scope you need."

  Priya's response, recorded because it is the common one:
  "But then who does it?"
  -> the answer had to be concrete (the rotation, by name, with
     dates) before she would let go of it.

"But then who does it" is the reliable response, and a vague answer means nothing changes, because the person doing glue work stops only when they can see it is genuinely covered.

Results, one year on:

                              before        after
glue work share, most-loaded
  person                        ~60%          ~22%
interesting work to former
  close colleagues            4 of 6      2 of 9, then even
lead's share of the most
  interesting work               40%          ~10%
Priya's promotion             not in cycle  promoted, on the
                                            streaming migration
                                            plus the onboarding
                                            and retro artifacts
regretted attrition               n/a           0

The person who had also applied for the lead role was promoted
to senior the same cycle, on the plan built from that first
awkward conversation.

And one thing that did not resolve cleanly:

One of the two close friendships did become more distant, and
the lead's honest assessment was that this was partly
unavoidable and partly their own overcorrection. It was raised
by the friend, months later, using the permission granted in
the week-one conversation.

The lead's recorded note: "I was harder on him in reviews than
I would have been on anyone else, to prove to myself I wasn't
favouring him. He noticed before I did."

Recorded because the alternative reading, that the transition
conversation solves this, is not true. It makes it discussable.

Production evidence

Tanya Reilly's "Being Glue" talk and essay is the canonical treatment: it names the category, argues that the work is essential and non-promotable, documents that it accretes on the most conscientious people, and makes the case that leaders must both distribute it and value it explicitly rather than relying on gratitude.

Research on non-promotable tasks (Babcock, Recalde, Vesterlund and Weingart's work on the "office housework" pattern, published in the American Economic Review and summarised in The No Club) finds that such tasks are disproportionately assigned to and volunteered for by women, and that the effect persists in randomised settings, which is why this is a fairness issue rather than only a career one.

Google's re:Work materials on manager transitions and the broader literature on first-time managers consistently identify the peer-to-manager transition as the highest-failure-rate transition, with the two documented failure modes being avoidance of difficult conversations and overcorrection into distance.

Camille Fournier's The Manager's Path and Michael Lopp's writing both treat the explicit transition conversation with former peers as a required early action rather than an optional courtesy, on the grounds that the relationship has already changed and the only question is whether it is renegotiated deliberately.

Published engineering career ladders express advancement in terms of scope and artifacts, which is the structural reason glue work does not promote: it produces outcomes without artifacts, and the calibration room reads artifacts.

Proximity bias research in hybrid teams documents the analogous invisibility problem for remote workers, and the remedy is the same: measure the distribution rather than trusting a manager's recollection of who did what.

The debate

Should you keep any technical work as a lead? Some, and the useful boundary is that you should keep work that nobody's growth depends on: the tedious migration, the on-call shift, the boring but necessary refactor. What you must not keep is the most interesting piece, because that is the developmental one, and the plausible story ("this is the risky part") is always available.

Is the transition conversation awkward enough to skip? It is awkward for ten minutes and the alternative is awkward for a year. The specific value is the closing question, "is there anything about this that's awkward", which only produces information if asked in week one before there is an incident it could be about.

Can you stay friends with a former peer you now manage? Differently, yes, and pretending nothing changed is the failure. The overcorrection is the more common and more damaging error among conscientious people, and it is only correctable if you named the risk early and gave them permission to raise it, because you will not notice it yourself.

Should glue work simply be eliminated? Some of it, yes, and the audit usually finds a recurring meeting or a report nobody can justify. Most of it is genuinely necessary, which is why rotation plus artifact conversion is the answer rather than deletion. The version that fails is telling people to stop doing it without saying who does it instead.

Is converting glue work to artifacts just gaming the promotion system? The written version is genuinely more valuable: a written onboarding path serves every future hire where a helpful colleague serves one, and a quarterly retro pattern analysis found a repeated root cause that one-retro-at-a-time had missed. The artifact is better work, and being promotable is a side effect rather than the purpose.

Should you tell someone their glue work will not promote them? Yes, explicitly, and the framing matters: the work is valuable, the reward system does not see it, and managing that gap is the lead's job. Leaving it unsaid means the person discovers it at calibration, which is both worse and later.

Follow-up Q&A

"What actually changes when you start managing former peers?"

You hold information you cannot share, so silence now means something; your casual opinion is a directive whether you intend it or not; complaining to you becomes a risk for them, so you stop hearing the complaints that used to reach you; and the friendship has an asymmetry it did not have. None of that is optional, so the only question is whether it gets renegotiated deliberately. The conversation is four points, said individually to everyone in the first two weeks, and it ends with "is there anything about this that's awkward," which only produces an honest answer if you ask before there is a specific incident it could be about.

"What is the most common failure of a first-time technical lead?"

Keeping the interesting work, with a plausible justification. It is the risky part, or you are fastest at it, or it touches the payment path. Every time, you take the most developmental piece away from someone who needed it and signal that the interesting work is not available to them. In one case the lead had kept two of the five most interesting items in a quarter, 40 percent, and only found out because a report said "you took the rate-limiter rewrite, that was the piece I wanted." The boundary that works is keeping work nobody's growth depends on.

"How do you avoid favouritism, and what is the trap in the other direction?"

Measure the assignments rather than trusting your read, because you cannot see your own pattern. In one first month, four of six interesting items went to the two former close colleagues, and the lead's contemporaneous read was that they were simply the right people. The trap in the other direction is overcorrection: being harder on your friend than the same behaviour would earn from anyone else, to prove impartiality to yourself. It is the more common error among conscientious people, it is unfair in a way that is harder to see, and you will not notice it, which is why naming it in week one and giving them permission to raise it is the only real mitigation.

"What is glue work and why does it matter to a lead?"

The coordination, communication and maintenance work that makes a team function and that no rubric rewards: onboarding, retros, the answer desk, runbooks, noticing the cross-team gap, chasing the dependency. It is high-value work with no attribution mechanism, and it accretes on whoever is most conscientious, which is not randomly distributed: the research on non-promotable tasks finds it falls disproportionately on women, and it persists in randomised settings. So it is simultaneously a team health issue, a career issue and a fairness issue, and the lead is the only person positioned to see the distribution.

"How do you fix it?"

Three things, all of them needed. Distribute it, with named rotations so the allocation mechanism stops being "whoever cares most," and delete the parts nobody can justify. Make it visible by converting it into artifacts, since a written onboarding path with measured outcomes is promotable where being helpful is not, and the written version is genuinely more valuable anyway. And protect the promotion path explicitly: tell the person you see the work, that it will not promote them on its own, that this is a failure of the system rather than of them, and that you are rotating half of it away and giving them named technical scope instead. Expect "but then who does it," and have a concrete answer with names and dates, because a vague one means nothing changes.

"How do you argue for glue work in a calibration room?"

Convert it into scope evidence with numbers. Not "they're the glue of the team," which is an adjective, but "they built the onboarding path that took time-to-first-commit from 19 days to 3, ran the retro programme whose action-item completion rate is 84 percent against an org average of 40, and caught two cross-team design mismatches before they shipped, one of which the platform team's own review missed." Same work, and the second version is evidence a room can act on.

Common misconceptions

"If I don't change anything, the friendship stays the same." It already changed on day one. Not naming it means it gets renegotiated by a series of small incidents instead.

"Being fair means treating my friend the same as everyone." It means treating them the same, which includes not being harder on them to prove something to yourself. That is the error conscientious people actually make.

"I should keep the risky work." The risky work is usually also the interesting work, and that is the developmental piece. Keep the tedious necessary things instead.

"Glue work is low-value work." It is high-value work with no attribution mechanism, which is a different problem with a different fix.

"The people doing glue work should just stop." They stop only when they can see who does it instead, by name and date. Telling them to stop without that changes nothing except how guilty they feel.

"Recognising glue work in a review is enough." Recognition without redistribution keeps the same person doing 60 percent of it next quarter, and recognition without an artifact does not survive calibration.

Interview delivery note

Say this verbatim: "Glue work is essential, invisible and non-promotable, and it accretes on whoever is most conscientious rather than being assigned. So the lead's job is three things: rotate it so the allocation stops being self-selection, convert it into artifacts so it can be credited, and tell the person explicitly that it will not promote them on its own, because they usually think it is just what a good colleague does." Naming glue work unprompted is a strong people-leadership signal, and the three responses show you have actually done something about it.

The senior-versus-staff separator is measuring the distribution rather than trusting your read. A senior lead notices that one person does a lot of the invisible work. A staff lead runs a half-hour quarterly audit listing everything that happened that was not on the roadmap, counts it per person, finds one person carrying 60 percent, and discovers that their own prior read of that person as "steady but not stretching" was a misreading of someone doing most of the team's uncredited work. The same discipline applied to assignment found four of six interesting items going to two former close colleagues in a month the lead thought was fine.

The second signal is naming the overcorrection risk out loud in week one. Saying "I'm going to be careful about fairness in a way that might feel like distance, and if you ever think I'm treating you harder than others to prove a point, tell me, because I might be and I won't notice" shows you know the more common failure among conscientious new leads is distance rather than favouritism, and that you cannot detect it yourself.

The Staff & Team Lead Interview Reference