The fastest way to slow an AI build is to be the person whose calendar every decision routes through. Most six-to-twelve-week MVP engagements ship late not because the partner missed a milestone, but because the founder unintentionally became the queue. The pattern is reliable — same four behaviors, same engagement-weeks, same week-3 surfacing — and it is fixable with four artifacts the founder writes in about two hours at week 0.
The founder-AI-partner operating manual frames the week-by-week ritual; the idea-to-product manifesto frames the broader operating model. This piece is the mechanic-level cut: where the founder shows up as the bottleneck and what artifact replaces the behavior.
Thesis: the bottleneck is a throughput problem
The 2026 baseline AI MVP engagement is six to twelve weeks, two to three engineers, a fixed-price or capped-time-and-materials SOW, and an eval-gated definition of done. Within that shape, the partner runs roughly fifteen-to-thirty decisions per week — model choice, retry policy, whether a test result counts as a regression, ship-to-staging timing. Most do not need the founder. A small fraction does.
The bottleneck appears when a founder routes the wrong fraction through themselves — either too many decisions (everything queues on their approval) or the wrong ones (strategic decisions slip past while operational ones get scrutiny). The McKinsey 2024 State of AI survey reports roughly 30% of organisations cite “lack of clear strategy” as the top AI adoption barrier; at MVP scale that usually does not mean the SOW is unclear — it means the founder is operating against the SOW in week 3 and the partner cannot say so. Bain’s 2024 Generative AI report reaches a similar conclusion at enterprise scale: roughly 40% of pilots stall on governance and decision rights, not technology.
The corrective is not “be more responsive”. A founder who answers every Slack ping inside an hour but cannot articulate which decisions they own is still the bottleneck — they have made the queue invisible. The corrective is to make the boundary explicit at week 0 so the partner can run their weekly decisions without round-tripping to a calendar, and so the founder gets the right two-to-four every week with enough context to decide on the spot.
Week-cost figures below are illustrative heuristics drawn from typical six-to-twelve-week engagements, not specific deployment statistics.
Behavior 1: approval gating non-strategic decisions
What it looks like. The partner asks in Slack: “For the document parser, do we want a retry on OCR, max two attempts?” The founder reads it Tuesday, defers because they are unsure, responds Thursday with “sounds fine.” Three similar pings sit in the queue. The partner has stopped work on the parser because they cannot land a structurally trivial change without approval.
Week-cost. Roughly half a build-day per gated decision when the partner blocks; eight gated decisions across an engagement compound to one to two build-days of slip.
Partner-side symptom. Standup updates contain “waiting on founder for the retry pattern”, “parked the parser thread, will resume when we hear back.”
The fix. Pre-approved decision lanes (Fix 1). The OCR retry sits in the partner-decides lane. The founder hears about it in the weekly sync as a one-line note, not as an inbound approval.
Behavior 2: re-litigating closed decisions
What it looks like. Week 1: founder and partner agree the MVP supports English only, French in v2. Week 4: a customer call surfaces a French request. The founder messages: “Should we maybe add French now?” The partner re-opens the architecture, re-scopes the eval set, re-prices the build. By the time the founder says “actually let’s stick with English,” roughly a build-day is spent and English-path momentum has dropped.
Week-cost. Half-to-one build-day per re-litigation. Three to five across an engagement compound to one to two weeks of slip and a sharp drop in partner trust.
Partner-side symptom. The technical lead’s tone shifts from confident to careful. They start adding “as discussed last week” and “per the SOW” to threads, and ask for written confirmation before acting on Slack messages.
The fix. The written decision record (Fix 2). The week-1 English-only decision is in the record with a timestamp and a rationale. A week-4 customer call goes into a new row labelled “reconsider English-only?” — and the default is to keep the original decision unless new evidence clears a documented bar.
Behavior 3: silent disapproval
What it looks like. Week 3 demo. The partner shows the parser working end-to-end. The founder smiles, says “looks great,” and the meeting moves on. Inside their head: “the UI is wrong, the way errors are surfaced is wrong, and I do not understand why we chose two-attempt retry instead of three.” None of it gets said. By week 6 the founder has accumulated twelve silent reservations. They surface as a Slack message: “I think we need to step back and rethink the parsing flow.” The partner reads it as a pivot from a happy founder.
Week-cost. The most expensive of the four per occurrence. A silent reservation surfacing in week 6 typically costs one to two weeks of rework, because architecture was locked on the assumption that week-3 sign-off was real. Three silent disagreements can compound to a four-week slip.
Partner-side symptom. The partner does not see it in real time — that is what makes it expensive. Late symptoms include the founder’s weekly engagement dropping and demo enthusiasm flattening.
The fix. The weekly 30-minute decision sync (Fix 3) with a structural slot for “things I am not sure I agree with” — a calendared, normalized place to surface disagreement before it accumulates.
Behavior 4: absentee weeks
What it looks like. Week 4. The founder is on a fundraise sprint or a customer trip. They miss the Monday standup, the Wednesday eval review, and the Friday demo. They reply to Slack at midnight with one-line answers. No designated alternate has been named. The partner is making strategic-shaped decisions on the founder’s behalf without authority to commit.
Week-cost. Roughly one build-week per absentee week without decision-lane coverage. The cost is not “no work happens” — work happens in the lowest-risk subset of the backlog while critical-path items quietly slide.
Partner-side symptom. Standup updates contain “working on the secondary backlog while we wait,” “will defer the auth flow decision until [founder] is back,” “making provisional choices on UI copy that may need rework.” Cumulative risk surfaces in the next demo as quiet anxiety about scope.
The fix. The designated alternate (Fix 4) — a named human (cofounder, advisor, or domain-savvy contractor) who holds founder-equivalent decision authority within pre-agreed boundaries during absentee weeks.
Fix 1: pre-approved decision lanes
A one-page artifact written at week 0 that splits every decision the partner will encounter into three lanes: partner-decides, founder-decides, needs both. Categories, not cases.
| Lane | Examples |
|---|---|
| Partner decides — informs in weekly sync | Model choice within budget, retry/timeout policy, library choices, internal API shape, test fixtures, deployment timing within the sprint, refactor decisions, code-style conventions |
| Founder decides — partner can recommend | Product surface and customer-facing copy, pricing, brand voice, MVP-done rubric, which persona the eval set represents, go/no-go on customer-visible launches |
| Needs both — synchronous sync slot | Scope changes adding ≥1 build-day, eval-rubric changes, pivots away from documented week-0 assumptions, vendor lock-in (model provider, datastore class), pricing-model changes that affect architecture |
How to write it. Walk the SOW deliverables in the week-0 kickoff. The partner names the categories of decisions they expect to face; the founder marks each one. Disagreements get discussed on the spot — that is the point. Treat the output as a living but heavy document — moving a category mid-engagement is a meaningful change, not a Slack message.
Prevents. Approval gating (behavior 1) and, partially, absentee weeks (behavior 4) — because the partner-decides lane keeps the build moving when the founder is offline.
Trade-off. Founders sometimes worry this cedes too much. It cedes operational authority while consolidating strategic authority — the founder sees and decides every needs-both item, and the partner-decides lane is observable, not invisible.
Fix 2: the written decision record
A single document — Notion page, Google Doc, Slack canvas — that holds every decision made in the engagement, with three columns: decision, rationale, date. New rows append; existing rows do not get edited (changes go in a new row referencing the old).
| Decision | Rationale | Date |
|---|---|---|
| MVP language scope: English only; French in v2 | Customer interviews showed 92% of pipeline is English; French infra adds ~4 build-days and a French eval set | 2026-W01 |
| OCR retry policy: max 2 attempts | Three customer documents had transient failure modes that cleared on second attempt; third adds latency without recovery | 2026-W02 |
| Eval rubric for parser: 85% field-extraction accuracy on 50-document holdout | Industry baseline; aligns with downstream-process tolerance per customer interview | 2026-W02 |
| Reconsider English-only? Closed: keep English-only | Week-4 customer call requested French; pipeline weighting unchanged; cost of v1 French still ≥4 days | 2026-W04 |
How to write it. First entries go in at week 0 from the SOW and kickoff. Every weekly decision sync produces 1–5 new rows. The partner appends, the founder reviews, and the record is the canonical answer to “what did we decide?” — not Slack history, not memory, not the SOW.
Prevents. Re-litigation (behavior 2). A founder messaging “should we add French?” in week 4 is told: “we have a decision row from week 1 — what new evidence clears the bar?” The conversation either produces a new row (decision reversed with named reason) or closes (decision stands).
Trade-off. A few minutes per decision. Obvious win by week four.
Fix 3: the weekly 30-minute decision sync
A standing 30-minute meeting, once a week, with a fixed agenda: needs-both decisions queued since last sync, partner-decides items worth informing on, an explicit slot for the founder’s “things I am not sure I agree with.” This is not the demo. This is not standup. This is the decision meeting.
| Slot | Minutes | Content |
|---|---|---|
| Needs-both queue | 15 | Walk the 2–4 needs-both items the partner flagged since last week. Decide on the spot; capture in the decision record. |
| Partner-decides recap | 5 | Partner names the 5–10 partner-decides decisions taken since last week. Founder confirms no surprises. |
| Founder uncertainty slot | 5 | Founder names 1–3 things they are not sure they agree with. Surface only; resolve in follow-up if non-trivial. |
| Next-week preview | 5 | Partner names decisions likely to land in the coming week. Founder flags any expected absence (Fix 4 trigger). |
How to run it. Calendar it before week 1. Keep it 30 minutes. Show up. The first three weeks feel slightly mechanical; by week 4 it is the meeting the founder protects.
Prevents. Silent disapproval (behavior 3), by giving disagreement a normalized place to land. Also catches approval-gating drift (behavior 1) by surfacing whether the partner is queueing too many “ask the founder” items.
Trade-off. Thirty minutes per week is the cheapest insurance against four-week rework. Skipping this is the single most common shortcut founders take, and the one most likely to cost a renegotiation.
Fix 4: the designated alternate
A named human, agreed at week 0, who holds founder-equivalent decision authority within pre-agreed boundaries when the founder is unavailable for more than two business days. Named, not implied. Written, not inferred.
| Field | Example |
|---|---|
| Name | Pat Q., advisor and former PM |
| Domain | Product surface, copy, customer-facing decisions |
| Boundary | Authority on needs-both items the founder has not pre-decided; cannot reverse decisions in the written record; cannot approve scope changes >2 build-days |
| Trigger | Founder unavailable >2 business days (auto-notification to partner and alternate) |
| Out-of-scope | Pricing, hiring, fundraise context, strategic pivots — these always wait for the founder |
How to set it up. In kickoff, name the human. Get their consent. Share the boundary document with the partner. If no human is available, the founder commits to a calendar lane during absent weeks — e.g., one 30-minute Zoom every other day — to keep the queue moving. “No alternate” is itself a choice, but it has to be a named choice with a compensating block.
Prevents. Absentee weeks (behavior 4) — the build does not stall in week 4 because Pat can decide the needs-both UI copy and the partner can keep moving.
Trade-off. Requires investing trust in another person and writing down the boundary of that trust. Small investment; one-to-two weeks of build velocity recovered.
Putting it together: a week-0 setup
| Artifact | Owner | Write at | Maintain at |
|---|---|---|---|
| Pre-approved decision lanes | Founder + partner together | Week 0 kickoff | Reviewed quarterly or on lane-change event |
| Written decision record | Partner maintains, founder reviews | Week 0 (seed entries) | Appended after every decision |
| Weekly 30-minute decision sync | Founder | Calendared in week 0 | Standing, 30 mins, weekly |
| Designated alternate | Founder | Week 0 (named human, agreed boundary) | Activates on founder-absence trigger |
Total founder-time investment is roughly two hours at week 0 — one to write the lanes, one to identify and brief the alternate — plus thirty minutes per week thereafter. Against a six-to-twelve-week engagement that ships on time instead of three weeks late, the return is straightforwardly positive.
For founders pre-kickoff, the AI MVP Scoping Worksheet is a one-page template that establishes the week-0 lanes, the decision record’s seed rows, and the alternate’s boundaries before the build starts — so week 3 finds the artifacts in place rather than scrambling to create them after the bottleneck is already visible.
Decision scope
This article addresses founders inside, or about to enter, a six-to-twelve-week AI MVP engagement with a chosen partner. It does not address founder behavior with an internal engineering team (covered in the founders-first-90-days playbook and the weekly cadence guide), partner-selection criteria (covered in the best idea-to-product partners for solo founders), or founder behavior pre-engagement, where briefing and scoping mechanics matter more than throughput.
The four-behavior taxonomy is observed across roughly the same engagement-week windows: approval gating from week 1, re-litigation from week 3, absentee weeks from week 4, silent disapproval surfacing by week 5–6. The fixes are sequenced for that ordering — lanes and record at week 0, sync starting week 1, alternate activated on absence trigger.
Frequently Asked Questions
How is the founder-bottleneck different from the five founder anti-patterns? The five founder anti-patterns are the diagnostic — what is going wrong and what it looks like. This piece is the prescription — what four artifacts the founder runs themselves. The anti-patterns piece is week-cost by behavior; this piece is fix-by-fix. Use one to diagnose; use this one to repair.
Is the weekly decision sync the same as the weekly demo? No. The weekly demo and cadence is partner-driven and progress-oriented. The decision sync is symmetric and decision-oriented. Combine them only if the engagement is small (one engineer, four-week build); otherwise keep them 24 hours apart on the calendar — demo Friday, decision sync Monday or Tuesday.
What if I do not have a candidate for designated alternate? Name the absence explicitly. Tell the partner: “I do not have an alternate; during absent weeks, the needs-both queue parks.” Then commit to one calendar block per absent day for decision-clearing. Worse than having an alternate, better than implicit “we’ll figure it out” — which is the absentee-week default and the costliest version.
How do I write decision lanes if I do not know what decisions the partner will face? Walk the SOW deliverables. For each deliverable, ask: “name the five decisions you expect to make on this — what are they and who decides each?” The partner has seen the shape before; you bring the domain context. Lanes get written collaboratively in kickoff, not in advance by either party.
Can a great partner save a founder who runs none of these artifacts? Partially. A disciplined partner will surface the patterns explicitly — naming approval-gating in standup, refusing to act on re-litigation messages without a written request, escalating absentee weeks. But the artifacts only work if the founder owns them; a partner cannot run a decision record for a founder who does not write decisions down.
Does this work if my partner is offshore or asynchronous? Better, in fact. Decision lanes, the decision record, and the alternate’s written boundaries map cleanly to async work — the weekly sync is the only synchronous artifact, and 30 minutes weekly is sustainable across most time zones. Offshore engagements that fail tend to fail on implicit decision rights, not on time-zone mechanics.
How heavy does the decision record need to be? Light. Three columns — decision, rationale, date. No status, no owner, no JIRA-style apparatus. If a row takes more than two minutes, it is too heavy and the team will stop using it. Heavy decision records die; light ones compound.
What if my partner refuses to use the artifacts? A serious signal. A partner that resists pre-approved decision lanes, a written decision record, or a 30-minute weekly sync is either inexperienced with founder-engagements or planning to operate in ambiguity. Raise it explicitly in week 1 — and if it does not resolve, reconsider the engagement before week 3.
How does this compare to RACI or other decision-rights frameworks? RACI is a heavy artifact for internal teams. Decision lanes are the lightweight founder-partner adaptation — three lanes, not four roles, category-based, not task-based. Inside a six-to-twelve-week engagement, RACI is over-fitted; decision lanes are the right altitude.
Closing
The founder-bottleneck pattern is the single most common reason AI MVP engagements ship two-to-four weeks late, and it is fixed by four artifacts the founder writes in roughly two hours at week 0. Approval gating, re-litigation, silent disapproval, and absentee weeks have specific operational counterparts: decision lanes, decision record, weekly sync, designated alternate. None of these are heavy, and the compound effect is a build that runs at the partner’s natural cadence rather than at the founder’s calendar.
For the week-0 template version, the AI MVP Scoping Worksheet is the one-page download that puts the lanes, the record’s seed rows, and the alternate’s boundaries on paper before the build starts.
Arthur Wandzel
