AI Decision Frameworks: Pre-Mortem, Red Team, Devil's Advocate and SWOT — Run by Agents
Frameworks are useful when they change the questions a team asks. They are harmful when they become decorative labels on the same comfortable answer.
A pre-mortem, red team review, devil's advocate, and SWOT analysis each create a different pressure on a decision. The pre-mortem imagines failure from the future. The red team attacks the plan as an opponent. The devil's advocate argues against the preferred option. SWOT maps internal and external conditions, but needs a decision rule to avoid becoming a four-box inventory.
AI agents make these frameworks more repeatable by assigning each job to a specialised role, running the work against the same context, and preserving the arguments for synthesis. The value is not that an agent is magically objective. The value is that the process makes dissent harder to skip.
TL;DR
- Use a pre-mortem to expose plausible failure modes before committing resources.
- Use a red team to model how a competitor, market, or hostile stakeholder could defeat the plan.
- Use a devil's advocate to challenge the preferred option and reveal motivated reasoning.
- Use SWOT as a map of conditions, then convert it into choices, tests, and owners.
- Ask agents to execute the frameworks as distinct jobs, not as a single prompt pretending to be four people.
- For a focused decision process, see the Decision Sprint offer. Thornfield Partners' 12 partners adopted the brief format and averaged under one hour to review a completed brief.
1. Pre-mortem: assume the decision failed
A pre-mortem starts with a future fact: it is 12 months later and the decision has gone badly. The team must explain why.
This is different from asking, “What risks do you see?” That question invites a list of generic caveats. The pre-mortem creates psychological permission to be specific. It asks people to describe the failure as if it already happened: the enterprise launch missed the pipeline target, the senior hire left after six months, or the acquisition absorbed the product team without creating growth.
Use it when the team is excited, aligned, and moving quickly. Enthusiasm is precisely when weak assumptions are least likely to be challenged.
How an agent executes it: The risk agent receives the decision brief and writes five to ten plausible failure narratives. The operations agent translates each narrative into an early warning signal. The finance agent estimates exposure and runway impact. The synthesiser ranks failures by likelihood, severity, and detectability.
Sample prompt:
It is 12 months after we approved this decision and the outcome is
unambiguously bad. Write six specific failure narratives. For each:
name the broken assumption, the earliest observable signal, the owner
who could detect it, and one mitigation that does not depend on luck.
The output should not be a risk register full of “market conditions may change.” It should say what changed, how the team would notice, and what action follows.
2. Red team: attack the plan as an opponent
A red team does not ask whether the plan is sensible from the inside. It asks how the plan could be defeated by an intelligent competitor, a sceptical buyer, a constrained team, or an adverse market.
Use a red team for market entry, positioning, security-sensitive launches, acquisitions, pricing changes, and any plan where the external response is part of the thesis. The red team should have permission to make the strongest case against the plan, not merely to add a risk paragraph after the recommendation is written.
How an agent executes it: The strategist defines the plan's winning conditions. The customer agent models why a buyer would not switch. The competitive agent models the fastest credible response. The operations agent tests whether the team can respond. The final report separates threats that require a plan change from threats that are acceptable trade-offs.
Sample prompt:
Act as a well-funded competitor with a faster sales cycle. We are
launching the plan below. Identify the three moves you would make to
make it fail, the assumption each move exploits, and the evidence that
would tell us you are doing it. Do not suggest generic differentiation.
The best red-team output is uncomfortable and actionable. If it only says “competition could increase,” the team has not red-teamed the plan.
3. Devil's advocate: challenge the preferred option
The devil's advocate is narrower than a red team. It does not attack from the outside; it argues that the decision-maker's preferred option is wrong, premature, or framed too generously.
Use it when the room has already converged, when a founder has a visible emotional preference, or when the cost of being wrong is asymmetric. The role is especially valuable after a long research process, when the amount of evidence can create confidence without creating clarity.
How an agent executes it: The contrarian agent identifies the preferred option from the brief, reconstructs the strongest case for the leading alternative, and names the assumption that would have to be true for the preference to be correct. The finance and customer agents then check whether the argument changes the economics or only the rhetoric.
Sample prompt:
We currently prefer Option A. Make the strongest honest case for
Option B. Identify what we are emotionally protecting in Option A,
which evidence we are overweighting, and the one test that could
justify changing the recommendation.
Devil's advocacy is not a requirement to choose the opposite option. It is a requirement to earn the preferred option.
4. SWOT: map conditions into decisions
SWOT is familiar because it is easy to start. Strengths and weaknesses describe internal conditions; opportunities and threats describe external conditions. The failure mode is stopping at the grid. A list of adjectives is not strategy.
Use SWOT at the beginning of an unfamiliar decision to create a common map, or when a team has different definitions of the situation. Do not use it as the only framework for a high-stakes choice.
How an agent executes it: The strategist drafts the map from the supplied evidence. The finance agent challenges strengths that do not produce economic value. The customer agent tests whether a claimed strength is visible to buyers. The risk agent turns threats into scenarios. The operator converts the grid into actions, owners, and a review date.
Sample prompt:
Build a SWOT for this decision using only the supplied facts. Mark
each item as observed, inferred, or unknown. Then convert the two most
important items in each quadrant into a choice, experiment, or
monitoring metric. Do not leave the output as a four-box list.
The labels “observed,” “inferred,” and “unknown” matter. SWOT becomes misleading when assumptions are written with the same certainty as evidence.
How to combine the frameworks
The frameworks are strongest as a sequence, not a checklist. Start with SWOT to map the context. Use the devil's advocate to challenge the leading option. Run a red team to model external defeat. Finish with a pre-mortem to identify the internal failure modes and early warnings.
An AI council can distribute those jobs across agents, then ask a final synthesiser to answer:
- Which objections change the decision?
- Which objections change only the implementation?
- What is the smallest test that resolves the highest-impact unknown?
- What will we monitor, who owns it, and when do we review?
- What dissent should remain visible in the final brief?
The last question is a quality test. If the synthesis removes every disagreement, the process has produced consensus theatre. A good Decision Audit Trail records the recommendation and the reasoning that nearly changed it.
That is the pattern Thornfield Partners adopted across 12 partners. The shared brief made the review faster — under one hour on average — because everyone could see the decision, the challenge, and the unresolved question before the discussion began. The human meeting remained essential; it simply started later in the process.
Choosing the right framework
| Situation | Best first framework | What it should produce |
|---|---|---|
| The team is optimistic and moving fast | Pre-mortem | Failure narratives, signals, mitigations |
| Competitor or stakeholder response matters | Red team | Attack paths and defensive moves |
| The room has already converged | Devil's advocate | Strongest alternative and mind-changing test |
| The situation is unfamiliar or disputed | SWOT | Shared map converted into choices and experiments |
Run more than one framework when the decision is expensive to reverse. For a small reversible decision, one focused challenge is usually enough.
Frequently asked
Can one AI prompt run all four frameworks? It can produce four labelled sections, but labels do not create independent challenge. Better results come from assigning distinct objectives, letting agents inspect the same brief, and requiring the synthesis to show what changed.
Which framework should I use first? Start with the question you are avoiding. If you are afraid the plan will fail, run a pre-mortem. If you are afraid a competitor will respond, red-team it. If you are attached to one option, use a devil's advocate.
Is SWOT too basic for strategic decisions? SWOT is basic as a map, not as a decision process. It becomes useful when each item is sourced, uncertainty is marked, and the output leads to tests, choices, owners, and review dates.
Can an AI framework replace expert judgment? No. Frameworks structure inquiry; humans still supply accountability, context, values, and decisions that affect other people. Use the output to improve a board, advisor, or leadership conversation, not to avoid one.