The Weekly Founder-Partner Cadence: what good looks like

A non-engineer founder running an AI MVP engagement spends about 90 minutes a week with the partner team — or the engagement decays. Less than that and the founder becomes the absentee who shows up at week 6 with surprise notes. More than that and the founder becomes the over-meddler whose engineers spend more time defending decisions than making them. The 90-minute version is a specific shape: three 30-minute meetings wrapped in a thin async layer of one Loom, one Slack channel, and one decision log. Each meeting closes one decision loop. That is what a good weekly cadence looks like.

This is a working framework that fits into Founder Operating with an AI Partner, part of the idea-to-product manifesto. That guide sketches the 12-week shape; this piece zooms into the weekly rhythm that runs from week 2 through week 12.

Thesis: the cadence is the engagement

Two AI MVP engagements with identical SOWs, identical staffing, and identical scopes can produce wildly different outcomes — the variance is almost entirely on the cadence axis. The partner shows up either way. The founder shows up in one of two failure modes or one working mode, and the cadence selects which.

The first failure mode is the absentee founder: signs the SOW, ghosts for three weeks, shows up at the demo with surprise notes — “I assumed this would handle the Spanish-speaking segment” — and discovers the eval suite was scoped to English and the partner shipped what was specified. The second is the over-meddler: reads every commit, second-guesses the architecture in week two, and ends week three with engineers who have spent more time defending decisions than making them. Both ship a technically-impressive demo that is commercially off-target.

The working mode is the operating-rhythm founder: three 30-minute meetings a week, one async Loom reviewed on their own time, one written scope decision per Friday, engineers otherwise left alone. It is a calendar, not a personality trait. A founder without this rhythm can run it from Google Calendar with three recurring blocks and a one-page artifact checklist. The 90-minute budget is deliberate: below it the founder cannot stay current; above it the founder is in engineering decisions, which is a different (and almost always destructive) intervention.

The 90-minute weekly shape at a glance

Day	Meeting	Length	Lead	Decision output
Monday	Eval review	30 min	Founder	One scope decision on which rubric dimensions to defend, soften, or escalate this week
Wednesday	Demo + scope check	30 min	Partner	Written PR commitment list for Friday + any de-scope decisions logged
Friday	Customer review	30 min	Founder	New inputs added to the eval seed set + go/no-go on next week’s customer interviews

Plus a thin async layer: one weekly Loom posted by the partner every Wednesday by noon, one shared Slack channel (and only one), and one append-only decision log. Three meetings, three loops, three artifacts. The founder spends 90 minutes of synchronous time plus roughly 60 minutes of async review per week; the partner contributes whatever the SOW specifies (typically 60–120 engineering hours).

Monday — eval review (30 min, founder-led)

The single most important meeting of the week. Monday morning, 30 minutes, founder plus partner technical lead plus one or two engineers. The founder runs it. Fixed agenda:

1. Open the eval dashboard. Two minutes. The partner shares screen; the dashboard shows pass-rates across each rubric dimension for the last seven days, with deltas against the prior week and the defined floor.
2. Walk three to five inputs where scores moved. Twenty minutes. The partner picks the most informative movements — up or down — and walks each input live. The founder watches the model output and calls whether each movement is a real quality change or a measurement artifact.
3. Surface threshold drift. Three minutes. The partner names any rubric dimension where the team has informally moved the floor down to keep regressions from blocking merges. A team shipping at 88% against a 92% floor that quietly became 86% is in trouble — Monday is when that gets named.
4. Decision output. Five minutes. The founder makes one written decision: which rubric dimensions to defend, which to soften (with rationale), which to escalate. The decision goes into the log before the meeting ends.

Founder questions (use verbatim):

• “Has the score genuinely moved, or did we change what we are measuring?”
• “Are any rubric floors lower than they were two weeks ago — and when did we move them?”

The founder is the domain authority. The partner measures; only the founder can say whether a four-point improvement matters for the customer. The mechanics of the rubric itself are unpacked in the role of evals in your weekly partner relationship; the cadence point is that this meeting happens every Monday at the same time, with the same agenda, until the product ships. If three Mondays in a row produce no written decision, the cadence has decayed.

Wednesday — demo + scope check (30 min, partner-led)

The only partner-led meeting. Mid-week timing is deliberate — Monday surfaced the issues, Tuesday was for resolution, and the founder sees the result before Friday locks in.

1. Live demo of the week’s work. Ten minutes. Not slides — actual product surface. The partner walks the founder through what changed in the user-facing experience since last Wednesday.
2. Scope check against the rubric. Ten minutes. The partner names any scope decision needing a founder call — a dimension harder to ship than estimated, a feature the eval data suggests re-prioritizing, a customer signal from Monday warranting re-scope. The founder decides in the meeting or names a deadline.
3. PR commitments. Five minutes. The partner names the specific pull requests committing to land by Friday. PRs, not story points — PRs are countable. The companion piece on the 30-minute standup covers the partner-side daily mechanics that produce these commitments.
4. Decision output. Five minutes. Any scope change goes into the log; PR commitments are also recorded — they are the contract the partner is making about Friday.

Founder questions:

• “Which PR on this list is the one most likely not to land — and what is the contingency?”
• “Is the rubric still right, or has anything we learned this week changed priorities?”

Wednesday is not a status meeting. Status is delivered via the asynchronous Loom; Wednesday is reserved for two-way decisions that cannot wait until Friday and cannot be made on Slack. If three Wednesdays in a row produce no scope decision, ask explicitly: “What is the hardest scope call you are about to make this week?”

Friday — customer review (30 min, founder-led)

The most-skipped meeting in AI MVP engagements, and the one that most clearly distinguishes a working cadence from a ceremonial one. Most cadences end the week with a status review; the working cadence ends with a customer review — the founder bringing the partner fresh signal from the people who will pay for the product.

1. The week’s customer evidence. Ten minutes. The founder shares the customer-conversation note — at least one conversation with a target user, ideally three. New phrasing, new objections, new edge cases. The partner listens and asks clarifying questions.
2. Eval seed-set updates. Fifteen minutes. The founder and partner together turn the customer evidence into new inputs for the seed set — five to fifteen new representative inputs, written as if they were real user messages. Committed before the meeting ends.
3. Go/no-go on next week’s customer activity. Five minutes. The founder names the customer interviews or product checks happening next week. The partner names any feature dependency — e.g., “we need the Spanish-language fallback live by Tuesday for the prospect call Wednesday.”

Founder questions:

• “What did a customer say this week that would surprise the engineering team?”
• “Is anything the customer wants outside the current rubric — and should it be?”

Friday is the founder’s primary contribution. Customer evidence is the one artifact the partner cannot manufacture; without it, the rubric drifts toward whatever is easy to measure rather than what matters to a buyer. The companion piece the founder’s role in an AI MVP build covers the broader split of work. If three Fridays produce no new seed-set inputs, the founder is operating in absentee mode.

The async layer

The 90 synchronous minutes are bracketed by three async elements.

The weekly Loom demo. Every Wednesday by noon the partner posts a five- to ten-minute walkthrough of the week’s shipped work — replacing the synchronous status meeting most agencies default to. Loom-first decouples founder review from partner build time, lets the founder rewind confusing sections, and produces a permanent, forwardable archive. The structure: sixty seconds of context, five to eight minutes against the eval suite or staging, one to two minutes on next week with PR-level specificity, thirty seconds of links. The studio-side version is unpacked in inside the SFAI Labs operating cadence.

One Slack channel — and only one. Shared between the founder and the full partner team. Three rules: no DMs about engagement work; no urgent decisions in long threads (if it cannot wait until Wednesday, it goes on a 15-minute call); async by default. The single-channel rule is the most-violated and the most important.

The decision log. A single document where every scope decision and rubric change is recorded as a one-paragraph entry: date, decision, rationale, who decided, what changes downstream. Append-only. The log is what prevents the week-9 dispute about who agreed to what in week 3. A working log has one to three entries per week — fewer and the meetings are not producing decisions; more than five and the engagement has too much scope motion.

The three anti-cadences a founder should refuse

The cadence above is what good looks like. Three cadence shapes look reasonable in a kickoff conversation and produce predictable damage by week four. A founder who recognizes them on sight can refuse them before they ossify.

Anti-cadence 1: the daily founder standup. The partner offers a daily 15-minute standup with the founder included as “client representative.” It sounds attentive; it is structurally destructive — 75 minutes a week spent listening to engineering blockers the founder cannot resolve. The founder either zones out (training the team to ignore the founder) or starts intervening on engineering decisions (the over-meddling failure mode). A founder belongs in zero daily standups.

Anti-cadence 2: the open-ended weekly Slack thread. “We’ll just stay in touch on Slack” as the primary cadence. No fixed meetings, no fixed artifacts. Maximizes optics in week one, produces total information loss by week four. Slack threads are not meaningfully searchable after two weeks. Decisions get made implicitly, unmade implicitly, with no log to reconcile against. By week six no one remembers when the Spanish-language requirement was de-scoped.

Anti-cadence 3: the bi-weekly “how are you feeling” check-in. A 60-minute every-other-week relationship-temperature meeting in lieu of a weekly eval review. The structural problem: two weeks is too long between eval reviews — a rubric regression on Tuesday of week three can ship to production on Thursday of week four because no one is looking. AI MVPs require weekly eval cadence because model behavior, prompts, and rubric thresholds drift on a weekly timescale. The 10 rules of working with a partner — including this one — are unpacked in the 10 rules of working with an AI agency without losing leverage.

What good looks like by week 4

By end of week 4 a healthy cadence has produced specific, countable evidence:

• Twelve meetings held. Four Mondays, four Wednesdays, four Fridays. No skips, or a written explanation for any skip.
• Six to twelve decision-log entries — one to three per week, each with date, decision, rationale, decider, downstream change.
• Four weekly Looms posted, four watched with the founder’s written feedback comment on each.
• A growing eval seed set, roughly doubled from week-1 baseline, traceable to specific Friday conversations.
• Zero unresolved threshold-drift notes — floors held, or every drift event has a written rationale.
• One Slack channel, capped DMs — no parallel DMs about engagement work.

A founder with all six is running a healthy engagement and can predict, with reasonable accuracy, what week 6’s demo will show. A founder with fewer than four is in trouble — the gap is the leading indicator of how off-target the week-6 demo will be.

Frequently asked questions

What is the weekly founder-partner cadence in plain English?

A 90-minute-per-week buyer-side operating rhythm for an AI MVP engagement: three 30-minute meetings (Monday eval review, Wednesday demo and scope check, Friday customer review) plus a thin async layer of one weekly Loom, one shared Slack channel, and one append-only decision log. Each meeting produces one specific written decision. The founder runs Monday and Friday; the partner runs Wednesday.

Why is the eval review on Monday rather than Friday?

Monday positions the eval signal at the start of the partner’s work week so they have Tuesday through Thursday to act on it. A Friday eval review surfaces regressions only after the week has shipped — the regression then sits over the weekend and through Monday before anyone works on it.

Does the founder need to attend the partner’s daily standup?

No. A partner who invites the founder to daily standups is proposing one of the three anti-cadences. The daily standup is an internal-team rhythm covering blockers the founder cannot resolve. The founder gets the same information, condensed, via Wednesday’s demo, the weekly Loom, and the Slack channel.

What if the partner refuses to record a weekly Loom?

Push back. If the refusal holds, treat it as a structural signal — a partner who cannot produce a five- to ten-minute weekly Loom is either understaffed, disorganized, or trying to keep the engagement under-documented to preserve optionality at exit.

What does the decision log actually look like?

Append-only one-paragraph entries: date, decision (one sentence), rationale (two to three sentences), who decided, what changes downstream. Example: “2026-06-17. Decision: deprioritize Spanish-language fallback for v1; ship in week 8. Rationale: this week’s customer evidence confirmed the pilot cohort is all English-speaking; engineering cost is four engineer-days and is better spent on multi-document retrieval. Decided by founder. Downstream: rubric dimension D7 paused; D2 gets the engineering time.”

How do I run the Monday eval review if I am not technical?

You do not need to be technical. The partner provides the data and screen-share; you provide the customer judgment. The exact phrases in the agenda — “Has the score genuinely moved, or did we change what we are measuring?” — are designed to be used verbatim by a non-engineer founder. Run the agenda mechanically for three weeks and you will be running it fluently by week four.

What happens if I miss a week of meetings?

One week is recoverable — record what happened in the next log entry and ask the partner to compress two weeks of demo into one Wednesday session. Two weeks is a structural signal. Three weeks is engagement-altering; the recovery is not “more meetings” but a sit-down about whether the engagement should continue and on what terms.

How is this cadence different from agile or scrum?

The unit of commitment is the pull request, not the story point — PRs are countable and demo-able. There is a dedicated weekly meeting for the eval suite, which scrum does not have because non-AI work does not need it. And the founder is an operator of the cadence rather than a stakeholder in it.

When does the cadence stop?

When the engagement ends — with one caveat: a working cadence keeps the Friday customer review for roughly 30 days after launch, even if Monday and Wednesday drop to bi-weekly. The post-launch 30 days are when the founder verifies the product is solving the problem in production.

Closing

The weekly founder-partner cadence is 90 synchronous minutes, three named meetings, three written decision artifacts, one Loom, one Slack channel, one log. It is not glamorous. It is the boring, mechanical, repeatable substrate that turns a 6-to-12-week AI MVP engagement into a product that ships on target. The next move for a founder reading this is to download the AI MVP Scoping Worksheet from the founder operating manual and use it to draft the rubric Monday’s eval review will operate against from week 2 onward.