IGNITE EXTENDED / LESSON 03
What working agreements help a team make and execute better decisions?
A one-page working agreement with an escalation example.
A team can hold a productive discussion and still leave without a decision because members assume different decision rules. One person may expect consensus while another expects the project lead to decide after consultation. The resulting frustration comes from an unspoken process disagreement.
A working agreement should identify who contributes information, who makes the decision and how unresolved objections are recorded. Use a rule appropriate to the consequence and reversibility of the choice. A reversible experiment often needs a clear owner more than unanimous enthusiasm.
Task conflict concerns the explanation, assumption or proposed action. Relationship conflict concerns perceived motives, respect or personal standing. A disagreement about a forecast can become personal when people infer that questioning the forecast means questioning their competence.
Make the disputed statement observable. Ask which input differs, what each person expects and what evidence could discriminate between the explanations. A shared decision criterion makes it easier to compare options. When evidence remains incomplete, record the uncertainty and a review trigger rather than disguising disagreement as agreement.
These are different responsibilities in a meeting and can be performed by one person or shared. Making the ownership explicit prevents the team from assuming that someone else will close the work.
Chooses the purpose and process of a discussion, makes space for evidence, and closes with decisions.
Tracks contributions, deadlines and delivery. Coordination does not mean deciding every substantive issue.
Illustrative application: the research lead guides a discussion about segment choice and ensures competing evidence is heard. The coordinator records that recruitment depends on a contact list due Tuesday. Good facilitation without that handoff can still leave the experiment unable to start.
The next process joins the discussion to a decision and then to observable commitments.
A meeting is useful when it changes a decision or coordinates work. Consider a team choosing which customer segment to test next week.
Send the decision, alternatives and evidence needed. Ask research to bring interview observations and sales to bring access constraints. State who will decide and which issues remain outside this meeting.
Compare the same alternatives against the same criteria. Separate observed customer behavior from market-size estimates and personal enthusiasm. Ask which assumption explains each disagreement.
Apply the agreed rule. A named owner can choose a reversible experiment after hearing objections. Record a dissent that identifies a meaningful risk rather than demanding superficial agreement.
Read back the action, owner, output and due date. Example: research recruits five clinic administrators by Friday, then the group reviews whether the chosen segment can be reached.
The decision rule determines how discussion closes. The commitment determines what another person can rely on after the meeting.
A wheel of five rules for team meetings: set the rotation of meeting leaders in advance; each leader plans the format and agenda; each assignment has a coordinator, distinct from the leader, who handles coordination and submission; all members are equally responsible for meetings and assignments; some meetings are structured as part of learning.
The page separates leading a discussion from coordinating delivery and assigns each to a named person ahead of time. That separation is the root of this lesson's distinction between deciding and tracking: when nobody owns closure, work stalls even after good discussion.
The first four questions for a team's first meeting: get to know each other; what each person wants to learn from the work and from the team; what counts as a successful outcome for the task; and how the team defines a high-performing member.
Questions 3 and 4 make standards explicit before work begins. Disputes about effort or quality later are easier to resolve when the team has already agreed what 'successful' and 'high-performing' mean.
A promise to “help with research” leaves the recipient unable to plan. A usable commitment identifies an output, an owner, a due time and the condition for acceptance. It also describes dependencies that could prevent delivery.
The meeting leader can close a discussion by reading back decisions and commitments. This gives participants a final opportunity to identify incompatible expectations. Missed commitments should prompt an early conversation about the observable gap and its consequence, followed by a revised commitment or a change in ownership.
A team benefits from different expertise when members can combine it in a decision. Different backgrounds also produce different vocabulary, standards of evidence and definitions of completion. Expertise that remains isolated does not automatically improve the outcome.
Use a common artifact, such as a customer journey, model or decision memo, to make the contributions comparable. Ask each specialist to explain the consequence of a technical issue in terms of the shared objective. Preserve the technical detail in the supporting analysis so that simplification does not hide a material constraint.
The remaining questions: how the team will operate, built around four core norms (how decisions are made, how open members will be, how disagreements are handled, whether members will hold each other accountable when someone is not carrying their load); what each person brings to the team; and how each person tends to get in their own way and how they want others to respond.
The four norms map directly onto a working agreement: decision rights, openness, disagreement, accountability. Question 7 is unusual and useful: each person pre-authorizes how teammates should flag their known failure mode, which lowers the cost of raising it later.
A fictional team repeatedly discusses a launch date. An engineer sees reliability risk, a salesperson sees a customer deadline, and nobody knows who decides. Each meeting ends with “let’s think about it.”
Which process gaps should be repaired before another meeting?
Many remote and fast-growing companies formalize single-point ownership. GitLab's public handbook defines a DRI as the person ultimately accountable for a project, with final decision authority after consulting stakeholders; if a DRI cannot change a decision-maker's mind in reasonable time, they are expected to disagree and commit. Written records like this matter more when a team is distributed, as Relay's will be.
Paul Graham's September 2024 essay, prompted by a talk from Airbnb's Brian Chesky, argued that standard advice to hire good people and give them room can fail in founder-led companies, and that founders may need to stay involved in details. Graham also cautioned that delegation is still necessary and that the idea can be misused as an excuse not to delegate.
The debate is about where a founder's decision rights end, which is exactly what a working agreement should state and revisit as the team grows from 4 toward 22. Write the boundary down rather than leaving it to style.
Sources: GitLab Handbook: Directly Responsible Individuals (DRI) · Paul Graham: Founder Mode (September 2024)
Use the launch-date situation. The customer needs a decision by Friday. Reliability test results will be available Thursday. One person has missed two assigned research tasks.
Specify decision rights, the evidence cutoff, a disagreement rule and a response to a missed commitment. Solo learners compare two possible agreements.
Month 6. Relay v1 is live at 12 clinics (illustrative) at $399 a month; the pre-seed SAFE raised $1,500,000. Daniel wants to spend the next quarter building an AI assistant that converses with patients by text, arguing that LLMs make it feasible now and competitors will get there first. Maya wants engineering focused on making v1 sell: EHR integration reliability and faster onboarding, because two pilots stalled at setup. Their discussions keep ending without a decision, and Daniel has twice missed the date for an onboarding fix while prototyping the assistant.
Write a one-page founder working agreement: decision rights by decision type, the disagreement and escalation rule, how the build-AI-now question will be decided (evidence, deadline, review trigger), and how missed commitments are raised.
52/48 can feel like a tiebreaker, but using ownership to settle product disputes teaches the minority founder to disengage. Assign decision rights by decision type and expertise.
Ending the AI-assistant debate with 'let's revisit later' leaves both founders working to different priorities. Record the dissent and a specific result or date that reopens the question.
An agreement drafted at month 0 for two people will not fit a team of nine. Review it when the team or the stakes change, for example after the seed adds a board seat.
A one-page Relay founder working agreement with decision rights by type, the build-AI-now decision with recorded dissent and review trigger, a commitment format and an escalation path.
Next lesson: Empathy & customer evidence
Check your understanding
Question 1
Maya and Daniel own 52% and 48% of Relay. Daniel argues that because Maya's majority will always win votes, they should decide everything by consensus. What is the strongest response?
Operating decisions need a clear owner matched to expertise and consequence. Consensus on everything invites delay and hidden vetoes; 'majority owner decides all' ignores Daniel's technical authority and would push him out of decisions he is best placed to make.
Question 2
Maya and Daniel keep arguing about building the AI assistant now versus selling v1. Which restatement turns their argument into a productive task conflict?
It names each person's assumption and the evidence that discriminates between them. The first attacks motive and turns task conflict into relationship conflict; deferring without a trigger is a dominated delay; the last substitutes a preference for evidence.
Question 3
Relay's head of growth, Priya, keeps the meeting notes, tracks deadlines and chases overdue tasks. A new engineer assumes she therefore decides the product roadmap. What is the misunderstanding?
A coordinator closes loops on commitments; a decider owns a substantive choice. Conflating them either overloads the coordinator or leaves decisions unowned when the coordinator assumes someone else will decide.
Question 4
Relay must choose which of two pilot onboarding flows to try next week. The choice can be reversed after 10 clinics. Which decision rule fits best?
A cheap, reversible experiment needs speed and a clear owner more than unanimity. Consensus and board votes add delay disproportionate to the stakes; waiting for complete data is impossible when the experiment is how the data are produced.
Question 5
Relay's EHR reconnect fix has slipped twice and two pilot clinics are stuck at setup. Which commitment can a teammate actually rely on?
It names the output, owner, due time, acceptance condition and dependency. The others leave the recipient unable to plan, and the dependency line surfaces the risk of a miss before it happens.
Question 6
Daniel has missed two committed dates for the onboarding fix while prototyping the AI assistant. Which response best follows the working agreement?
Early, specific, private conversation addresses the observable gap and finds the cause (here, likely competing priorities). Public blame converts it into relationship conflict; silence lets the pattern continue; banning the prototype treats the symptom without agreeing priorities.
Workbook
Module output: A decision journal, a short explanation and a working agreement.
Sources
Original pages and full documents are in the Library edition.