2026-10-29 · ~22 min read · Decision records

The engineering 1-on-1 cadence decision record: why the cadence model you chose determines your team health visibility ceiling and your retention risk detection latency

Engineering 1-on-1 cadence decisions are not choices about how often managers meet with their direct reports or which template to use. They are a set of founding decisions about which topics the dedicated manager-engineer conversation must cover to be a reliable signal source rather than a scheduling habit, how often the cadence needs to repeat to produce a meaningful trailing window of team health signal, how insights from individual 1-on-1s are documented in a form that persists beyond the conversation and can be used to detect behavioral change over time, how the career development conversation is integrated into the cadence in a way that produces durable commitments rather than aspirational discussions that are never revisited, and how the skip-level architecture provides signal about the quality and coverage of the direct manager's 1-on-1 practice before a retention event surfaces the gap — decisions whose consequences are invisible when the engineering team is small enough that informal conversations create sufficient signal, and visible when an engineer resigns and the manager realizes they had no documented signal about the trajectory that produced the resignation.

A 22-person developer tools startup had been running for fourteen months when two of its five senior engineers submitted resignations within three weeks of each other, both citing variants of the same concern: they felt disconnected from the product direction and could not see a growth path in the current role. The engineering manager was surprised by both resignations. Neither engineer had raised the concern directly. Looking back through Slack logs and sprint retrospective notes after the fact, the manager could identify the signals: one engineer had stopped asking questions about the roadmap in team meetings two months earlier; the other had mentioned feeling "a bit stuck" in a casual hallway conversation six weeks earlier and had not brought it up again. The manager had registered both signals as low-weight data points and had not followed up in a dedicated conversation. The startup had no formal 1-on-1 cadence. The manager met informally with direct reports as needs arose — before a difficult sprint, during a code review discussion, after a retrospective — and had assumed that the accessibility of these informal touchpoints would create sufficient signal about team health. The informal touchpoints did create signal. The signal existed in casual messages and verbal asides. The model had no mechanism for amplifying the low-weight signal ("feeling a bit stuck") into a dedicated conversation where the underlying concern could be fully articulated and addressed before it accumulated into a resignation decision. When both engineers left, the startup lost seven months of institutional knowledge about a proprietary data pipeline and four months of accumulated context about a customer integration that had been in progress. Neither engineer had a successor. Both had mentioned, in exit interviews, that they had felt that their concerns were not important enough to schedule a dedicated conversation about. Connect this to the engineering knowledge management decision record: the institutional knowledge that left with those two engineers — the architectural decisions about the data pipeline, the context about why specific customer integration approaches had been tried and rejected — is precisely the class of knowledge most at risk when a retention system produces no early warning; the organization that detects a senior engineer's disengagement six weeks before resignation has time to initiate a knowledge transfer process; the organization that detects it in the exit interview does not.

The informal-touchpoints model fails not because informal conversations produce no signal but because it produces signal whose weight is systematically underestimated. An engineer who says "feeling a bit stuck" in a hallway conversation has surfaced a retention risk. They have not escalated it to the level of urgency that prompts a manager to schedule time to address it, because the informal context signals that the remark is low-stakes and can be revisited later. "Later" does not reliably happen without a structural mechanism that converts the low-weight informal signal into a dedicated conversation. The 1-on-1 cadence is that mechanism: a recurring calendar slot whose existence signals that there is always a "next time" within a predictable window, and whose agenda creates permission to surface the concern more fully than a hallway conversation allows. The absence of that mechanism does not prevent engineers from feeling disengaged; it prevents the disengagement from surfacing in a form the manager can act on before it becomes a resignation.

A 48-person B2B SaaS had implemented mandatory weekly 30-minute 1-on-1s for all engineering managers eight months before the analysis that follows. The mandate was explicit: every direct report gets a weekly 1-on-1, every 1-on-1 gets a shared agenda in the calendar invite, cancellations require a reschedule within the same week. Compliance was high for the first three months — managers scheduled the meetings, engineers attended, agendas were created. By month four, the pattern had shifted. The agenda documents were mostly empty; managers opened the meetings by asking "what are you working on this week?" and the conversations tracked directly onto project status for the first twenty minutes. The remaining ten minutes were used for quick acknowledgments ("that sounds good," "let me know if that unblocks") and scheduling logistics. Engineers who had initially brought career and engagement topics to the meetings had stopped doing so — the conversational pattern had established that the agenda was project status, and raising a career question in that context felt like a non-sequitur. Participation remained technically high (the meetings happened), but eight months in, an engineer-experience survey found that 71% of engineers felt their career development was not being actively supported by their manager, and 58% said they would not bring a career concern to a 1-on-1 because "that's not what they're for." The mandate had achieved high cadence and near-zero signal on the topics that retention risk is actually made of.

The failure mode here is distinct from the startup's failure mode. The startup had no cadence and therefore no mechanism for signal to surface. The B2B SaaS had a cadence and therefore had a mechanism — but the mechanism had been captured by the least valuable use case. Project status has a strong pull in 1-on-1 conversations because it is always present, always specific, and feels productive to discuss — unlike career development questions, which require the manager to ask probing open-ended questions and create space for an answer that may not emerge cleanly. Status discussion is the path of least resistance in a thirty-minute slot, and without explicit structural guardrails against it, it tends to fill the available time. The ADR for the 1-on-1 cadence must address this not by prohibiting status discussion but by requiring explicit agenda coverage of non-status topics — by specifying that the career development question gets asked in every meeting, that the engagement signal question gets asked at minimum every other meeting, and that the manager's preparation before the meeting includes a review of the previous meeting's notes specifically to check whether any commitment made in the previous conversation has been followed up. Without those specifications, the mandate "run weekly 1-on-1s" produces weekly meetings, not the team health visibility that the mandate was intended to create. Connect this to the engineering performance management decision record: the performance review conversation that surfaces "this engineer has not been developing as expected" is frequently the first structured conversation the manager has about that engineer's trajectory — not because the manager was inattentive but because the 1-on-1 cadence that was nominally covering that topic had been running on project status for eight months; the performance management system and the 1-on-1 cadence must be explicitly coupled in the ADR so that the quarterly or biannual performance conversation is a summary of a trailing window of documented 1-on-1 signals rather than the first time the manager and engineer are discussing the trajectory that is being evaluated.

An 85-person engineering organization at a growth-stage enterprise SaaS had a 1-on-1 mandate that had been in place for eleven months: bi-weekly 1-on-1s required for all direct reports, no specified agenda format, no documentation requirement, skip-level conversations at the VP's discretion. The mandate was structured this way deliberately — the VP of Engineering believed that prescribing the conversation format would make 1-on-1s feel procedural and reduce the quality of the relationship. Each manager was trusted to run the conversations in the way that worked best for their team. Eleven months in, a performance review cycle surfaced a gap: six of the organization's twelve engineering managers had no documentation of career development conversations with direct reports — not in a formal system, not in notes, not in any form that could be reviewed by the manager's manager. When the VP asked those managers about their 1-on-1 practices, the responses varied substantially: two managers described weekly 45-minute conversations with structured agendas covering career development, team health, and technical progress; two described biweekly 20-minute meetings focused primarily on unblocking technical work; two described ad-hoc conversations that happened when needed, with the bi-weekly slot serving as a fallback that was often rescheduled or skipped. All six managers believed they were running effective 1-on-1s. The organization had no way to audit the quality of the conversations because no documentation requirement meant no retrievable signal. Engineers' survey responses about their 1-on-1 experiences showed a standard deviation more than twice as large as the industry benchmark the VP had been tracking — the org average was in the acceptable range, but masked a bimodal distribution where some engineers were receiving genuinely strong career development support and others were receiving periodic technical unblocking dressed as 1-on-1s. Connect this to the engineering manager span of control decision record: the calibration problem that the VP of Engineering encountered — twelve managers running 1-on-1s with structurally different quality levels and no common documentation model to compare against — is the same problem the span of control decision creates at scale; the manager of managers who is responsible for twelve managers' team health practices has no reliable signal about the quality of those practices without either a documentation model that surfaces the signal or a skip-level cadence that creates a direct line to the engineers receiving the practice; the span of control decision and the 1-on-1 documentation model are joint decisions whose consequences compound in either direction — a manager of managers with twelve reports and no documentation model is operationally blind to the quality variance in the team health system they are responsible for.

Structural properties set by the engineering 1-on-1 cadence decision

Two structural properties are determined by the founding decisions about cadence frequency, conversation coverage, documentation requirements, career development integration, and skip-level architecture. Both properties are invisible when the engineering team is small enough that informal relationships create sufficient signal, and visible when a retention event surfaces the gap between the conversations that were nominally happening and the conversations that were producing the signal the organization needed.

Property 1: Team health visibility ceiling. The team health visibility ceiling is the maximum signal available to the engineering organization about each engineer's engagement level, career trajectory, retention risk, and the structural conditions in the team that are affecting those outcomes. The ceiling is bounded by three factors. First, conversation depth: whether the 1-on-1 agenda covers the topics that produce signal about career development, engagement, and team friction — not just the project status that is already visible in the sprint board. A 1-on-1 that is structurally identical to a brief sprint review is not a signal source for team health; it is a check-in on work that would have been checked in on anyway, and its cancellation would produce no signal loss relative to the sprint board and async updates. The career development question ("Are you feeling like this project is moving you toward where you want to be?"), the engagement signal question ("Is there anything in the current environment that's making it harder for you to do your best work?"), and the working relationship calibration question ("Is there anything about how I'm managing you that you'd want me to do differently?") are the questions that produce signal available nowhere else in the manager-engineer relationship — but they are also the questions that are most easily crowded out by the pull of project status discussion and the discomfort of open-ended answers. The ADR must specify that these questions are asked, not leave them to the manager's discretion, because the manager's discretion under sprint pressure consistently de-prioritizes the uncomfortable question in favor of the comfortable one. Second, signal retention: whether the insights from 1-on-1 conversations are documented in a form that persists beyond the conversation. An undocumented 1-on-1 produces signal that lives only in the manager's memory, which means the effective signal retention is bounded by recall accuracy under the cognitive load of managing a full team while delivering on a sprint — a manager with six direct reports running bi-weekly 1-on-1s has twelve conversations per month, and the signal that "Alex seemed less energized than usual and mentioned the retrospective process felt performative" is unlikely to be recalled three weeks later when the next 1-on-1 arrives, which means the behavioral pattern that would have been visible across three consecutive conversations is invisible because the baseline is not recorded. A three-to-five line note after each 1-on-1 — topics raised, commitments made, observable engagement signals — creates the trailing window. Third, pattern recognition: whether the manager has a mechanism for identifying when an engineer's signals have changed over a trailing window of conversations, rather than assessing each conversation as a standalone event. The engineer whose engagement is declining will rarely surface the concern as a direct statement; it will appear as a pattern of changed responses to open-ended questions — shorter answers, deflected future-oriented questions, increased focus on process friction — that is only visible as a pattern across multiple conversations. A manager who reviews the preceding three 1-on-1 notes before each meeting has a different signal than a manager who enters each meeting with only the preceding week in memory. The team health visibility ceiling is not a function of the manager's empathy or relational skill; it is a function of whether the cadence model creates the structural conditions for empathy and relational skill to produce actionable, documented signal.

Property 2: Retention risk detection latency. The retention risk detection latency is the time between an engineer beginning to disengage and the organization having actionable signal about the disengagement — signal that is specific enough to act on, early enough that the intervention options have not been foreclosed by resignation momentum. The latency is determined by three factors. First, cadence frequency: how often the dedicated feedback space exists. Retention risk that goes undetected for a month becomes resignation momentum that takes three months to reverse; retention risk detected within two weeks after the first signal is actionable with a single career conversation, a project adjustment, or a promotion timeline clarification. The detection window is bounded above by the cadence frequency — monthly 1-on-1s set a floor of one-month detection latency for the signals that only surface in 1-on-1 conversations; bi-weekly 1-on-1s set a two-week floor. The frequency decision is therefore also a detection latency decision, and the ADR should be explicit about what detection latency the organization is accepting at each cadence level. Second, question model: whether the conversation structure within each meeting creates explicit space for the engineer to surface concerns that they would not raise in a public forum. The engineer who is considering leaving does not, in most cases, volunteer that information. They surface the conditions that are producing the consideration — the trajectory concern, the environment friction, the unmet expectation — in response to questions that explicitly invite reflection on those dimensions. "How is the sprint going?" is not that question. "Do you feel like the work you're doing now is building toward the seniority level you're aiming for?" is closer. "Is there anything about the current project allocation that's not sitting right with you?" is closer still. The question model in the ADR specifies the open-ended questions the manager is expected to rotate through the cadence, not as a script but as coverage requirements — the manager who has not asked a career trajectory question in three consecutive 1-on-1s has created a gap in the retention risk signal surface that is invisible to the org until an exit interview makes it visible. Third, escalation architecture: whether there is a path from a manager's 1-on-1 signal to a skip-level intervention when the signal indicates a retention risk that the direct manager relationship itself is not resolving. The class of retention risk that stems from friction in the manager relationship is the class that a manager-only 1-on-1 system cannot detect or address — the engineer who is considering leaving because of the manager's leadership approach, feedback style, or prioritization decisions will not surface that concern to the manager in a 1-on-1, because the manager is both the source of the problem and the person to whom the concern would need to be raised. A skip-level cadence — a regular conversation between the engineer and the manager's manager — creates a venue where this class of retention risk can surface. The escalation architecture also determines how quickly a high-severity retention signal that the direct manager has detected moves to the manager's manager for additional support — an engineering manager who has detected that a senior engineer is at retention risk should not be expected to resolve that risk alone with the tools available to an individual contributor manager; the VP of Engineering's involvement in a retention conversation can change the option set available (equity refresh, promotion acceleration, scope expansion) in ways that are not available to the direct manager acting independently.

The engineering 1-on-1 cadence ADR: five sections

Section 1: Cadence model and frequency calibration. The cadence model specifies the default meeting frequency, the meeting duration, the conditions under which the cadence is adjusted for individual engineers, and the rescheduling policy. The most consequential decision in this section is not the default frequency but the adjustment model: the conditions under which a manager appropriately changes an individual engineer's cadence, and the approval or documentation requirement for that change. An engineering manager who reduces a struggling engineer's 1-on-1 cadence from bi-weekly to monthly because the engineer expressed that the meetings were not valuable has made a structural decision to reduce the team health signal for that engineer — but this decision is invisible to the manager's manager unless there is a documentation requirement or an approval process. The adjustment model should specify at minimum: that cadence reductions require documentation of the rationale and the engineer's explicit agreement; that an engineer who expresses that 1-on-1s are not valuable receives a conversation about the format rather than a reduction in frequency; and that frequency changes are revisited after ninety days. The duration decision should specify the minimum viable duration for a 1-on-1 that covers all required topic categories — thirty minutes is sufficient for a bi-weekly meeting with a tenured engineer in a stable state; twenty minutes is not sufficient to cover career development, engagement signals, and working relationship calibration, and a meeting that consistently runs out of time before reaching career development has a structural coverage problem regardless of how often it happens. The rescheduling policy should specify that 1-on-1s are rescheduled within the same week when cancelled by either party, and that a pattern of cancellations by the manager (three or more in a trailing sixty-day window) is flagged in the documentation record as a potential signal about either the manager's workload or the manager's investment in the 1-on-1 practice. Connect this to the engineering team meeting cadence decision record: the 1-on-1 cadence and the team meeting cadence must be designed together to avoid the failure mode where a meeting-heavy schedule creates manager time pressure that produces systematic 1-on-1 cancellations — a manager running twelve weekly team meetings and ten bi-weekly 1-on-1s is already at thirty scheduled hours per month before any preparation time is counted, and the meetings that are cancelled under time pressure will be the ones that feel most optional, which in a culture that values output over relationship investment will be the 1-on-1s.

Section 2: Conversation structure and coverage model. The conversation structure specifies which topic categories must be covered in each 1-on-1 and at what frequency within the trailing cadence window, what proportion of the conversation should be engineer-led versus manager-led, and what topics are explicitly excluded from the 1-on-1 scope because they belong in other forums. The most consequential decision in this section is the exclusion model: which topics are explicitly redirected away from 1-on-1s. Project status that is visible in the sprint board belongs in the standup or in async written updates, not in a 30-minute 1-on-1 — but the manager who does not make this explicit will find that the status conversation fills available time because it is the path of least resistance for both parties. The exclusion model should specify that project status in 1-on-1s is limited to blockers that require manager intervention, escalations that require the manager to act before the next sprint planning, and context that is relevant to an ongoing career development discussion — not as a general-purpose status check. The coverage requirements should specify at minimum: a career trajectory question in every other meeting (more frequently during promotion cycles or periods of role change); an engagement signal question in every meeting ("Is there anything in the current environment that's making it harder for you to do your best work?" or an equivalent — the specific wording is less important than the category being covered); a working relationship calibration question once per quarter (less frequently for stable long-tenure relationships, more frequently during onboarding and the first six months of the manager-engineer relationship); and a manager observation and feedback component in every meeting, even if brief. The engineer-led time should be protected — the manager should not arrive at the 1-on-1 with a full agenda and leave no space for the engineer to raise topics that the manager has not anticipated. A useful structure is to begin the meeting by asking the engineer what is on their mind, reserve the first fifteen to twenty minutes for the engineer's agenda, and treat the manager's prepared topics as the second half — which means that if the engineer's topics require the full time, the manager's prepared feedback waits for the following meeting rather than truncating the engineer's space.

Section 3: Documentation and signal retention model. The documentation model specifies what is recorded after each 1-on-1, where it is stored, who has access to it, and how it is used in preparation for subsequent meetings and performance review cycles. The most consequential decision in this section is the access model: whether 1-on-1 notes are private to the manager, shared with the engineer, or visible to the manager's manager. Each choice has different signal properties. Manager-private notes produce maximum candor in the manager's documentation (the manager is more likely to record a candid observation about an engineer's engagement when that observation is not visible to the engineer) and minimum transparency (the engineer does not know what has been recorded about them, which creates a risk that the documentation is used in a performance cycle in a way the engineer did not expect). Shared notes produce a documentation dynamic where the manager records only observations they are comfortable sharing with the engineer, which filters out the most candid signal but creates a collaborative record that the engineer can add to and that surfaces disagreements between the manager's and the engineer's assessments of the same conversation. Manager's-manager-visible notes produce organizational signal about the quality and coverage of the direct manager's 1-on-1 practice, at the cost of reducing the manager's candor in documentation for sensitive topics. A workable approach for most organizations is a hybrid: the engineer can see a shared action item log (commitments made, topics flagged for follow-up), the manager's manager can see metadata (topics covered, cadence compliance, flag indicators for high-severity retention signals), and the manager's full observation notes are private. The pre-meeting preparation requirement should specify that the manager reviews the previous two to three 1-on-1 notes before each meeting to check for open commitments, behavioral change signals, and topics that were raised but not resolved — because the meeting that begins with "last time you mentioned feeling stuck on the growth path, I wanted to follow up on that" produces a fundamentally different quality of career development signal than the meeting that begins with "what are you working on this week?" Connect this to the engineering metrics decision record: the 1-on-1 documentation model determines which team health metrics the organization can produce — retention rate by manager, promotion rate by manager, engineer experience survey score variance by manager — that require a documentation record to calculate; an organization that tracks these metrics without a documentation model is measuring outcomes without any leading indicator signal, which means the first signal the metrics surface about a manager's 1-on-1 quality is the lagging indicator of retention loss or low experience survey scores rather than the leading indicator of cadence compliance and coverage that would allow intervention before the retention event occurs.

Section 4: Career development integration model. The career development integration model specifies how the 1-on-1 cadence connects to the formal career development and promotion process: how career conversations in 1-on-1s are documented in a form that persists into the performance review cycle, how the manager communicates the promotion criteria and trajectory to the engineer in a way that is specific enough to be actionable rather than aspirational, and how the commitments made in career development conversations are tracked and followed up. The most consequential decision in this section is the commitment tracking model: what happens to the commitments made in a career development conversation after the meeting ends. A manager who tells an engineer "I think you're on track for senior engineer in the next two performance cycles, and what would accelerate that is if you took more leadership on the upcoming infrastructure project" has made a career commitment that the engineer will remember and that the manager may not recall in the next performance cycle if the commitment was not documented. The commitment tracking model should specify that career commitments — specific statements about promotion trajectory, growth areas, stretch assignments offered — are documented in the shared action item log within 48 hours of the meeting, reviewed at the beginning of every third 1-on-1 to assess progress, and updated in the pre-performance-review preparation as the foundation for the formal evaluation. The promotion timeline communication model should specify how managers communicate promotion timelines in a way that is specific enough for the engineer to plan around: not "you're on a promotion track" but "based on your current trajectory, I'd expect to submit a promotion case for you in the next review cycle if the infrastructure project leadership goes as we've discussed, and here is specifically what 'going as we've discussed' means" — with a written artifact that the engineer can reference rather than a verbal commitment whose specifics are subject to memory variance on both sides. Connect this to the engineering career ladder decision record: the career development conversation in a 1-on-1 is only actionable if the career ladder is specific enough to support the conversation — a career ladder that describes seniority levels in terms of abstract qualities ("demonstrates leadership," "shows technical depth") does not give the manager and engineer a shared vocabulary for the career trajectory conversation; the manager who tries to run a specific career development conversation against an abstract career ladder will default to vague encouragement ("you're doing great") that produces no signal and no commitment, because there is no specific criterion to point to and no specific gap to name.

Section 5: Skip-level cadence and escalation architecture. The skip-level cadence model specifies how frequently the manager's manager meets individually with the manager's direct reports, what topics the skip-level conversation is designed to surface, and how signal from skip-level conversations is used to assess the quality of the direct manager's 1-on-1 practice. The most consequential decision in this section is the purpose model: what the skip-level conversation is for and what it is explicitly not for. A skip-level whose purpose is ambiguous produces one of two failure modes: engineers use it to escalate complaints about their manager (undermining the manager relationship), or engineers treat it as a performance review conducted by a more senior person (producing defensive rather than candid responses). A skip-level whose purpose is explicitly defined as "a venue for the engineer to surface anything about their role, their team, or their trajectory that they want the VP to know" — with an explicit statement that the skip-level is not an evaluation of the engineer and that the engineer will not be judged for raising concerns about their manager relationship — produces a different quality of conversation than an ambiguously-purposed meeting. The skip-level cadence for most organizations is quarterly for all engineers and monthly for engineers who are on active retention risk status — the manager's manager having quarterly signal about each engineer's experience is sufficient to detect the systematic patterns that direct manager 1-on-1s miss (manager-relationship friction, calibration disagreements about promotion criteria, organizational concerns that feel unsafe to raise with the direct manager), while not being so frequent that it creates an implicit secondary reporting line that undermines the direct manager relationship. The escalation architecture should specify how a high-severity retention signal moves from the direct manager to the manager's manager: what the signal threshold is that triggers escalation (an engineer who has explicitly raised a specific retention risk, an engineer whose 1-on-1 documentation shows a sustained behavioral change pattern over eight weeks or more), what the escalation looks like in practice (a conversation between the manager and their manager, not a formal HR process), and what the available intervention levers are at the manager's manager level versus the direct manager level — because the manager's manager arriving at an escalated retention conversation with only the same options the direct manager already has does not improve the retention outcome. Connect this to the open-source extractor: the reasoning that goes into a retention intervention — the manager working through with an AI assistant whether the signal they are observing constitutes a genuine retention risk or a temporary engagement dip, what intervention is appropriate given the engineer's career stage and the available options, whether to escalate to the skip-level now or wait for a clearer signal at the next 1-on-1 — is reasoning that is frequently conducted in AI chat sessions and is rarely preserved anywhere; the decision record that distinguishes "we recognized the signal, evaluated the options, and chose to offer a scope expansion, which the engineer accepted" from "we recognized the signal, the manager wasn't sure what to do, and the engineer resigned two weeks later" is made in that chat session, and recovering it is the organizational learning that determines whether the next manager in the same situation makes the same intervention or the same mistake.

FAQ

How often should an engineering manager run 1-on-1s with direct reports?

The right cadence depends on the team's current state and the purpose of the conversation, not on a universal frequency norm. For most engineering teams, bi-weekly 30-to-45-minute 1-on-1s are the minimum cadence that produces a useful trailing window of signal — frequent enough that behavioral change is detectable within a month, infrequent enough that each conversation has time to cover career development, engagement, and team friction rather than being consumed by project status. Weekly 1-on-1s make sense during significant change — a new role, a difficult project, a promotion cycle — because the feedback loop needs to be tighter to catch problems before they compound. Monthly 1-on-1s are too infrequent to detect retention risk at useful latency: a disengagement signal that emerges in week one and is not surfaced until week four is already month-old signal, and the engineer who was considering leaving in week one has had four weeks to solidify that decision before the manager has any signal. The cadence decision should specify different frequencies for different engineer states: a baseline cadence (bi-weekly for most engineers in a stable state), an active-development cadence (weekly for engineers on a promotion track or working through a significant role change), and a check-in cadence (monthly or as-needed for highly senior engineers who are satisfied in their current scope). Frequency is necessary but not sufficient — a weekly 1-on-1 that covers only sprint status produces less useful signal than a bi-weekly 1-on-1 that explicitly covers career trajectory and engagement; the cadence decision must specify both the frequency and the conversation structure.

What should engineering managers talk about in 1-on-1s?

Engineering 1-on-1s should cover the topics that are not covered anywhere else in the manager-engineer relationship — topics that require a private, dedicated space to surface reliably. Project status, sprint progress, and technical blockers have other venues (standups, sprint planning, design reviews, async updates); a 1-on-1 that allocates most of its time to these topics is replacing those venues rather than supplementing them. The topics that belong in 1-on-1s and have no other reliable venue are: career trajectory (where the engineer wants to be in one to three years, what skills they feel are missing from their current path, whether the current project is accelerating or stalling that trajectory); engagement and team health (whether there is anything in the team dynamics, the technical environment, or the organizational context that is making it harder for the engineer to do their best work); feedback in both directions (what the engineer wants more or less of from the manager, and what the manager has observed about the engineer's work that is worth discussing directly); and working relationship calibration (whether the manager's communication style and feedback approach are landing as intended). The split that most managers find useful is roughly one-third on the engineer's agenda, one-third on the manager's observations and feedback, and one-third on career development — not as a rigid formula but as a check against the pattern where one category consistently crowds out the others. Status discussion should be limited to blockers the manager can actually unblock in the next 48 hours; everything else visible in the sprint board does not need airtime in a 1-on-1.

How do engineering managers detect retention risk before an engineer resigns?

Detecting retention risk before resignation requires three capabilities that most engineering manager 1-on-1 practices do not systematically build: a trailing window of documented signal, awareness of behavioral indicators that precede resignation, and an escalation path for retention risk that originates in the manager relationship itself. The trailing window problem: a manager who does not document 1-on-1 notes has no mechanism for detecting that the engineer who was energized and curious three months ago is now consistently deflecting career questions — the change is invisible because the baseline has not been recorded. Documenting brief notes after each 1-on-1 (topics raised, commitments made, observable engagement signals) creates the trailing window that makes behavioral change detectable rather than retrospectively obvious. The behavioral indicators: the signals that precede resignation by four to eight weeks are usually visible in 1-on-1s before they surface anywhere else — reduced investment in future-oriented questions, increased focus on process friction, withdrawal from optional collaboration, and shorter answers to open-ended questions that previously produced extended discussion. None is individually conclusive; the pattern across three to four consecutive 1-on-1s is the signal. The escalation path problem: retention risk that stems from friction in the direct manager relationship will not surface in a manager-run 1-on-1 — it requires a skip-level mechanism where the engineer and the manager's manager have a regular conversation creating a venue where manager-relationship concerns can surface without the engineer needing to raise them directly to the person they are most concerned about.

How do 1-on-1 cadence decisions connect to AI chat sessions and decision preservation?

Engineering 1-on-1 cadence decisions are among the management decisions most frequently worked through in AI chat sessions — and among the decisions whose reasoning is most useful to preserve — because the decisions involve simultaneous reasoning across the organizational layer (what signal the team health system needs to produce), the interpersonal layer (what conversation structure creates space for the signals that matter most), and the systems layer (how individual conversations aggregate into organizational retention visibility). An engineering manager working through "I have a senior engineer who has been deflecting career questions for three months — what questions should I be asking?" with an AI assistant, a director working through "my 1-on-1 mandate is not producing consistent career development conversations across my managers — what is the structural fix?", or a VP working through "we lost two senior engineers in two months and the managers had no signal — what is the architectural problem?" — these sessions contain the reasoning that the calendar and the HR system do not record: what was considered and rejected as explanations for a behavioral shift, what retention lever was evaluated and why it was or was not used, what career conversation was attempted and how the engineer responded. Running those sessions through the open-source extractor produces the decision record that distinguishes a deliberate retention intervention (the manager diagnosed a trajectory mismatch, offered a project change, and the engineer re-engaged) from an undetected departure (the signals were present but the 1-on-1 structure had no mechanism for surfacing them) — two outcomes that look identical in exit interview data if the manager cannot reconstruct the decision sequence that preceded the resignation.

Structural properties set by the engineering 1-on-1 cadence decision

The engineering 1-on-1 cadence ADR: five sections

FAQ

Further reading