The Biweekly 1-1 Coaching Cycle
Every team member gets a biweekly 1-1 that runs as a loop, not a line. AI preps the conversation, a human holds it, and AI drafts a two-tier recap that reaches the member only after the coach reviews and publishes it. A mid-cycle check-in on the off week feeds the next prep, and a trend analysis refreshed after each 1-1 feeds the coaching focus. The AI preps and drafts; the human decides.
1-1s fail quietly, in five ways
Most leaders do not skip 1-1s. They run them without a system, and the same five things slip:
A loop with a gate, not a line
One person's two-week cycle. Solid lines are the working loop; dashed lines are the private layer's memory and the trend feedback. The recap has a review-then-publish branch, and nothing crosses from the private tier to the member without the coach opening the gate.
How each step works
The daily routine finds every 1-1 due within four days and writes the prep once: GROW-structured openers led by the goal question, a recommended mode, one thing to listen for, a retention check, and one question to avoid. It reads the person's goals, priorities, open commitments and last check-in updates, OCEAN profile, recent recaps, the coach's mode splits, and any talking points the member raised. The coach gets one digest per run, by Lark DM and email, with a portal link per prep.
The coach runs a GROW conversation with the mode chosen deliberately: coach (ask, and let them find it), mentor (share the experience), or direct (tell, because the situation demands it). The standing target is 80% coach, 15% mentor, 5% direct, and the system scores every meeting against it, so drift toward telling shows up as a number, not a feeling.
The meeting records through Lark Minutes in the background. The coach's job in the room is presence, not note-taking.
The daily routine auto-detects a new Lark Minutes recording whose title carries 1-1 and the member's first name, matches it to the meeting within a day, and pulls the transcript. If none is found, the coach pastes a Minutes link or raw text on the meeting. An hourly routine then drafts one held 1-1 at a time, oldest first, within the last 14 days, and tells the coach by Lark DM and email: a private summary for the coach (candid reads, retention signal, mode split estimate) and a shared recap for the member, plus the commitments made by either side. Always commitments, never tasks: they were agreed in a conversation, not assigned from above.
The coach edits both tiers; the shared recap reaches the member only when the coach publishes it.
Halfway through the cycle, if commitments are still open and no check-in has gone out since the last 1-1, the member gets one AI-written nudge, a Lark DM with an email twin, linking to the portal. Course corrections happen in week one, and every answer feeds the next prep: the loop talking to itself.
Once a person has two summarized 1-1s, the daily routine writes a private trend report, refreshed when the latest 1-1 has none for its month, and tells the coach by Lark DM and email: trajectory, recurring themes, follow-through, coaching opportunities, and flags worth raising. Patterns no one catches meeting to meeting become the next cycle's coaching focus. Every claim points to a behavior in the data; thin reads say low-confidence, and nothing is ever fabricated to fill a gap.
Goals, priorities, KPIs, commitments
Four concepts, four horizons, and they are never confused with each other:
| Concept | Horizon | What it is | Who sees it |
|---|---|---|---|
| FAST goals | Quarterly | What the person is driving this quarter: Frequent, Ambitious, Specific, Transparent. One or more per person, reviewed in every 1-1. | The whole team. Transparent is the T. |
| Priorities | Week to week | Standing focus items (P1, P2...) that open every 1-1. The working agenda between the goals and the day-to-day. | Coach + member |
| KPIs | Continuous | Numbers the person owns, with a target, a direction, and readings over time. A goal linked to one shows live progress automatically. | Coach + member |
| Commitments | Cycle to cycle | What was agreed in the 1-1, who owns it, by when. The member updates status and notes. Never called tasks. | Coach + member |
Goals ladder up ⇅ progress flows down
- Every FAST goal can ladder into the company goal tree at whichever altitude fits: an objective, a key result, or a KPI metric
- Coaching that moves a personal goal moves a company KR, visibly
- A metric-linked goal shows its live readings inside the 1-1 view
- The company-level goal health chips read the coaching goals directly, so "everyone has a current goal" is measured, not asserted
Why the distinction matters
- Goals are quarterly and public, so ambition is visible and peer-supported
- Priorities keep the 1-1 anchored without turning goals into a weekly status check
- KPIs stop progress debates: the number is on the screen
- Commitments make follow-through measurable without anyone playing task cop
What the coach knows, and who gets to see it
Two standing reads per person keep the coaching grounded, and both live behind explicit rules:
The OCEAN profile
- Five personality dimensions, each with a rating and behavioral evidence, plus a snapshot and growth guidance
- Coach-authored, never AI-published: the coach reviews before any profile goes live
- The member sees their own full profile once published, rewritten in second person so it reads as growth coaching, not a verdict
- Nobody else sees it: not the team, not the assistants
The retention read
- Built on job embeddedness: the three roots that keep people are belonging and fit, links to the people around them, and what they would sacrifice by leaving
- The AI names the thinnest root early, while it is still coachable
- Private to the coach, always: a retention read is a signal to act on, never a label on a person
- Most companies learn these answers in the exit interview, when they can no longer help
Privacy by access control, not discipline
The private coach layer and the person-facing layer are separate columns with separate query scopes. Mixing them is a code bug, not a human slip. The full access matrix:
| Data | Coach | Member | Whole team | AI assistants |
|---|---|---|---|---|
| FAST goals | Read + write | Read | Read | No access |
| Priorities + KPIs | Read + write | Read | No access | No access |
| OCEAN profile | Read + write | Read, own only, once published | No access | No access |
| Commitments | Read + write | Updates status + notes | No access | No access |
| Shared recaps + check-ins | Read + write | Read | No access | No access |
| Prep, transcript, private summary, mode split | Read + write | No access | No access | No access |
| Retention read + trend reports | Read + write | No access | No access | No access |
The assistant lockout is total: none of the coaching tables are readable by the company's internal AI assistants, the same treatment as sensitive HR data. Private coaching data never transits an assistant.
What runs it
| Tool | Role in the cycle |
|---|---|
| Two routines | The clock behind everything: a daily 07:45 run rolls next dates forward, writes preps, sends mid-cycle check-ins, refreshes trends, and auto-detects Minutes; an hourly run drafts one recap. |
| Lark DMs + email | Every nudge goes out twice: a Lark DM and an email twin, so nothing is missed. Links always land on the portal, where the data lives. |
| Lark Minutes | The transcript source. Recordings are matched to 1-1s by name in the title and date; the fallback is a paste. |
| Claude Opus 5 | Prep, recap drafting, and trend analysis. Every call fails soft: an AI error lands on the record and never blocks the meeting from happening. Every record stores the model that wrote it. |
| The team portal | The interface for both tiers: a coaching dashboard for the coach, a my-coaching view for the member, goals visible on team profiles. |
Why Claude Opus 5 writes the recaps
Every coaching generation, the prep, the recap, the check-in, and the per-person trend report, runs on Claude Opus 5. The model is named in exactly one place in our code, and every record stores which model produced it, so there is never a question about what wrote what. A full recap costs us under thirty cents.
| Model | Why, or why not |
|---|---|
| Claude Opus 5, what we run | The recap is the deliverable here, not a step towards it: a person reads it about their own work, and their coach acts on it. So we buy the model that reads half an hour of transcript most carefully. At one 1-1 per person per fortnight, the model bill is never the constraint. What the coach reads is. |
| Claude Sonnet 5, why not | Cheaper, and genuinely capable. We ran it head to head against Opus 5 on the same meeting with the same prompt: it returned the same sections with roughly half the detail. Saving eighteen cents on the one document the member actually reads is a bad trade. |
| Claude Fable 5.1, why not | Anthropic's most capable model, and more than this job needs. It costs twice Opus 5, and its reasoning cannot be turned down, so it spends tokens thinking about a task that does not need them and pushes against the output budget a recap has to fit inside. Summarising one honest conversation is not the hardest thing a model can do. We would rather spend the difference on more coaching than on more tokens. |
We re-check this choice rather than assume it. The model tier is a one line change, every run logs its own token cost, and a pilot can re-summarise a past 1-1 on a different model and put the record back untouched, so a comparison never costs anyone their history.
The AI preps and drafts. The human decides.
The rule that sits above every step: the AI never recommends promoting, managing out, ranking, or labeling a person. It surfaces the signal and proposes the move; the coach coaches the team. No permanent people decision is made during an emotional spike, and every claim the AI makes points to a behavior in the data.
The seven elements
Every workflow we document has the same anatomy: seven elements, each assigned to a human, a machine, or both. This is the Centaur Map from our workflow design method.
The calendar. A daily 07:45 routine rolls each profile’s next 1-1 forward by its cadence and keys prep, check-ins, and trends off it; an hourly routine drafts the recaps. Nothing waits on memory.
The person’s goals, priorities, KPIs, open commitments, OCEAN profile, and every prior 1-1, plus the transcript Lark Minutes hands over after the meeting.
Everything that touches a person: the mode to coach in, what the recap says, when it publishes, what the profile claims. The AI proposes; the coach decides.
Prep to the coach, check-ins to the member, every nudge as a Lark DM with an email twin, and each tier of the recap to its own access layer, enforced in code.
A prep doc per cycle, a two-tier recap, a living commitments log, and a trend report per person, refreshed after each summarized 1-1.
The team portal: a coaching dashboard on the coach’s side, a my-coaching view on the member’s side, and goals visible to the whole team.
Mode split per 1-1 against the 80/15/5 target, commitment follow-through rate, and goal progress against the company ladder.
The standing rules
- The 1-1 itself is the only human step, and it is the one that matters
- Nothing reaches the member without the coach reviewing and publishing it
- The two tiers never mix; the separation is enforced in code
- Commitments are agreed, never assigned, and never called tasks
- Thin reads say low-confidence; the AI never invents facts about people
- The coaching relationship is explicit and independent of the org chart, so any coach can run the same system
Why it works
- Every 1-1 gets the same depth of prep, on the busiest week and the quietest
- The check-in loop means nothing waits two weeks to surface
- The mode split turns "am I coaching or telling?" into a trendable number
- Transparent goals make ambition public and progress peer-visible
- Retention risk is named while it is still a coaching topic, not a resignation letter
- Roughly an hour of prep and recap per person per cycle comes back to the coach
This cycle is the operational half of a larger build. The program plan behind it, from problem statement to deployment, is published as The Leadership Coach: AI Program Plan.