A strategist who cannot reduce the agent program to six pages does not have a strategy. They have a backlog. The agent strategy memo is the artifact that converts a year of operating experience, a list of failed pilots, a portfolio of working agents, and a competitive landscape into a document the executive team can sign. The memo names the bets the program is making, the work it is killing, and the budget it needs for the next twelve months. It is not a manifesto. It is not a vision document. It is the operating contract between the agent program and the executives who fund it. A good memo lets the CFO defend the budget at the next board meeting, lets the CIO defend the program to the audit committee, lets the CEO answer the analyst question about AI strategy in one paragraph, and lets the agent program team push back on the next reorganization with a document everyone already signed. This lesson walks through the structure of a six-page memo, the one-page board summary that goes with it, the bets-and-kills framing that makes the memo decisional, the budget section, the failure modes that produce memos no executive will sign, and the rhythm by which the memo is updated. By the end you will have written a memo your CFO can read in fifteen minutes and approve in thirty.

Why The Memo Exists

The agent strategy memo solves a specific organizational problem. The agent program is an investment of millions of dollars across multiple teams, multiple vendors, and multiple risk surfaces. The executives funding it need to make decisions about it — to defend the budget, to align with broader strategy, to handle escalations, to make tradeoffs against other investments. They cannot make those decisions from a portfolio of dashboards and a quarterly review. They need a document that frames the program at the level executives operate at: bets, kills, budget, risks, and decisions.

What the memo replaces

Without the memo, the agent program is communicated to executives through three failure-prone channels. The QBR deck — too long, too tactical, optimized for the team that built it rather than the executive reading it. The board slide — too short, too abstract, useless for actual decisions. The hallway conversation — too dependent on individuals being available at the right moment, lost the moment people change roles. The memo replaces all three with a single artifact that exists at the right level of abstraction and persists across personnel changes.

What the memo enables

A signed memo enables four specific things. It enables the CFO to defend a budget line at the board with reference to a document the board has already implicitly approved. It enables the CIO to point a regulator or auditor to a single source of truth about what the program is doing. It enables the CEO to answer the analyst question — "what is your AI strategy" — in one paragraph that is consistent across earnings calls. And it enables the strategist running the program to push back on opportunistic asks ("can your team also build the marketing agent next month") by referencing a document that has the year's commitments written down and signed.

What the memo is not

The memo is not a vision document. It is not 30 pages of context-setting on the AI landscape. It is not a sales pitch for the agent program. It is not a list of every agent ever shipped. It is a six-page operating contract whose readers are executives who already know what an agent is and want to know what bets you are making with their money.

The test of a strategy memo is whether the CFO can read it in fifteen minutes, identify the three biggest line items, and explain back what the program is committing to. If the CFO cannot, the memo has failed.

The Six-Page Structure

The memo has six sections, each roughly one page. The pagination is real — if a section runs to two pages, the strategist has not done enough editing.

Page one: the state of the program

The current state in concrete terms. Not aspirational; not future-tense. Where the program is today.

This page covers: the number of agents in production (broken out by business unit), the cumulative production hours of agent runtime, the headline metrics across the portfolio (eval scores, escalation rates, cost per run, user satisfaction where measured), the three most consequential wins of the past twelve months in one sentence each, and the three most consequential failures in one sentence each. It also includes the program's current operating model (COE, federated, or hybrid) and headcount.

The state-of-the-program page is the baseline. Every other page in the memo is making a decision relative to this baseline.

Page two: the bets

The bets the program is making for the next twelve months. Each bet is a specific investment of money, headcount, and executive attention against a specific hypothesis about what will produce value. The bets are named, costed, and committed.

A bet is not "we will explore voice agents." A bet is "we will build a customer-service voice agent on Vapi, target call deflection of 30% on tier-one calls by Q3, invest $400K in build cost plus $150K in operating costs through year end, with a kill threshold of less than 15% deflection at end of Q2." Specific, datable, killable.

Three to five bets is the right number. Fewer than three and the program is not being ambitious enough; more than five and the executive team cannot remember them and the team cannot focus on them.

Page three: the kills

The work the program is killing or de-prioritizing. This page is structurally as important as the bets page; it is the page that signals operating discipline to the executive team. Most strategy memos have a bets page and skip the kills page; this is the mistake the lesson is designed to prevent.

Each kill is named, dated, and explained in one paragraph. Kills include: agents being sunset (per the kill plan from the previous chapter), agents being de-prioritized (not killed but not invested in further), capability investments being deferred (we considered building X; we have decided not to in 2026), and entire program directions being closed out (we are exiting the experiment with multi-agent orchestration on platform Y).

The kills page lets the executive team see the program is making hard tradeoffs, not just adding work. It also creates accountability — when an executive later asks why the program did not build X, the strategist can point at the kills page and say "we decided not to in May; here is the reasoning."

Page four: the budget

The budget for the next twelve months, broken out by category. Specific dollar amounts; specific allocation.

The budget categories typically include: foundation model spend (OpenAI, Anthropic, Google, Mistral, others), agent platform spend (Braintrust, LangSmith, Langfuse, Helicone, others), MCP server licensing and credentials, the platform team headcount (engineers, PMs), the agent-builder headcount in business units (or the cross-charge to those units), the shared-agents team headcount, training and enablement, professional services and vendor engagements, and the contingency line (typically 10-15% of total for unanticipated needs).

The budget is broken out two ways: by category (model spend, platform, headcount, etc.) and by bet (each named bet from page two has its budget allocation). The two views reconcile at the total.

Page five: the risks and the asks

The risks the program is carrying and the things the program needs from the executive team to manage them.

Risks are stated honestly. The top three to five risks: technical risks (the voice agent's latency profile may not work at scale), vendor risks (we are dependent on Anthropic's continued availability), regulatory risks (the EU AI Act enforcement details may require changes we have not budgeted for), organizational risks (we have not yet hired the platform team lead and time-to-hire is six months in this market), reputational risks (one customer-facing hallucination in the wrong context could damage the brand). Each risk gets a mitigation and an owner.

The asks are what the strategist needs the executive team to do. Approve the budget. Approve the headcount. Agree to the kill of the multi-agent investment. Provide the executive sponsor for the voice agent. Decide on the European data residency question. Asks are specific and have a date by which the strategist needs the decision.

Page six: the rhythm

The rhythm by which the program will report and decide for the next twelve months. This page commits the program to specific accountability mechanisms.

The rhythm typically includes: monthly executive update (one-page email with portfolio dashboard and any escalations), quarterly business review (full review of bets, kills, budget; this is where the memo gets updated), incident postmortem reviews (any P0 or P1 agent incident gets a postmortem reviewed at the next executive meeting), annual strategy memo refresh (this memo, regenerated annually), and the governance forum (the monthly or quarterly cross-functional meeting from the previous lesson).

This page also names who the strategy memo's signers are: the strategist (program lead), the executive sponsor (typically CIO or CTO), the CFO or financial controller, and any other necessary signatures (CISO if security risks are material, CPO if privacy risks are material, general counsel if regulatory risks are material).

The One-Page Board Summary

The six-page memo is for the executive team. The board sees a one-page summary derived from it. The summary is its own artifact, written specifically for an audience that has fifteen minutes and ten other items on the agenda.

The structure of the one-pager

The one-pager has four sections, each two to three paragraphs.

Where we are. A single paragraph: how many agents in production, what they cost, what they produce, what the headline outcomes have been. Numbers, not adjectives.

What we are doing next. One paragraph naming the three biggest bets. Just the names, the rationale, the cost, and the target outcome.

What we are killing. One paragraph naming the work being killed or deferred. This paragraph signals operating discipline — boards notice when programs admit what they are not doing.

What we need from the board. One paragraph or short bullet list of explicit asks: budget approval, headcount approval, any policy decisions the board needs to make. Asks have dates.

The graphics on the one-pager

The one-pager includes two small visualizations. The first is a portfolio table — every production agent in one row, with columns for owner, eval score, monthly cost, and status (running well, in remediation, sunsetting). The second is a budget summary — total budget for the year, broken into four to six categories with dollar amounts. The visualizations let the board reader see the program at a glance without parsing prose.

The discipline of the one-pager

A discipline more strategists violate than respect: the one-pager fits on one page. Not "one page if you use 9-point font." Not "one page plus an appendix." One page, readable font, with white space. The discipline of fitting on one page is the discipline of knowing what matters most.

The Bets Frame

The bets-and-kills framing is what makes the memo decisional rather than informational. A program without explicit bets has no commitments; a program with explicit bets has commitments and the executive team can hold the program to them.

What makes a bet a bet

A real bet has five properties: it is named (something a stakeholder can ask about by name), it has a hypothesis (what we believe will be true if the bet works), it has a quantified target (specific outcome at a specific date), it has a cost (specific dollar amount allocated), and it has a kill threshold (the specific condition under which we will stop investing). Without all five, the "bet" is an aspiration.

The three categories of bets

Most agent strategy memos in 2026 frame bets across three categories. Domain bets are bets on building a specific agent for a specific use case — the customer-service voice agent, the sales lead-enrichment agent, the finance close-acceleration agent. Capability bets are bets on building shared capability that multiple agents will benefit from — the eval platform, the guardrail library, the MCP registry, the platform team itself. Frontier bets are bets on emerging technology with high uncertainty but high potential — agent-to-agent protocols at scale, voice agents in regulated industries, on-device agents, the next-generation model the company is partnering on early.

A balanced memo has bets in all three categories. All domain bets and no capability bets means the program is not investing in the infrastructure that will let it scale. All capability bets and no domain bets means the program is building infrastructure for agents that do not yet exist. All frontier bets and no domain bets means the program is speculating instead of producing.

The bet sizes

The bet sizes should be unevenly distributed. The strategist who funds five bets at equal sizes is hedging, not betting. A typical distribution: one big bet (more than 30% of program budget), two medium bets (15-25% each), and one to two small bets (5-10% each). The big bet is what the program is most committed to; the small bets are options for the future. This distribution forces the strategist to actually decide what matters most.

The story of a betless memo

One mid-market software company's 2025 agent strategy memo listed eighteen "initiatives" with no explicit prioritization, no kill thresholds, no budget allocations against each. When the CFO asked which three the company was most committed to, the strategist could not answer cleanly. When the board asked what would be killed if the budget were cut by 20%, the answer was "we would have to think about it." The memo was not a strategy; it was a backlog with prose around it. The CFO cut the budget by 20% the next quarter; the program lost coherence; the strategist was replaced. The lesson: a memo without explicit bets cannot survive scrutiny.

The Kills Frame

The kills page is the page most strategists skip and the page that distinguishes a mature memo from an immature one. The reasons it is skipped — kills feel like admissions of failure, naming kills creates political controversy, the strategist is uncomfortable being explicit — are exactly the reasons it is essential.

Why the kills page signals maturity

Executives have seen a thousand programs that only added scope. They have seen far fewer programs that explicitly named what they were stopping. A program that names its kills is communicating three things: it has the operating discipline to make hard decisions, it understands that resources are finite, and it is honest about what is not working. All three are signals the executive team weights heavily because they are signals that the program is being run by adults.

The four types of kills

The kills page includes four types. Agent sunsets from the kill plan in the previous chapter — specific agents being decommissioned with their replacement paths. De-prioritized agents — agents that are still running but not being invested in further; explicit acknowledgment they will not get the eval refresh, the model upgrade, or the new feature this year. Capability investments deferred — things the program considered building (the bespoke eval platform, the multi-agent orchestration framework, the in-house guardrail layer) that have been explicitly killed or deferred. Program directions closed out — entire branches of the program tree that have been pruned (we are no longer pursuing the on-device agent strategy; we are no longer building agents for the marketing function until the marketing function asks for them).

The political handling of kills

Kills are politically sensitive because someone championed each piece of work that is being killed. The strategist's job is to handle the politics before the memo is finalized. Each named kill should have been discussed with the stakeholder whose work is being killed; the memo should not be the first time they hear it. The kills section also acknowledges the work that was done — "the X experiment produced valuable learning even though we are not productionizing it" — so that the people whose work is being killed see the work being respected, not erased.

The Budget Discipline

The budget page is where the memo becomes a contract. Vague budgets get cut; specific budgets get defended. The strategist must make the budget specific.

The level of detail

The budget is detailed enough that a financial controller can verify it, but summarized enough that an executive can read it. The right level of detail is roughly fifteen to twenty-five line items across two views (by category and by bet). More than thirty line items and the executive loses the forest; fewer than ten and the controller cannot verify it.

The unit economics

The budget includes the unit economics that let the executive team understand how the budget scales. Cost per run for each production agent (or the average across the portfolio). Cost per resolved support ticket for the support agent. Cost per qualified lead for the sales agent. The unit economics let the executive team translate budget to outcome: "we are spending $1.2M on the support agent because it is processing 4 million interactions a year at $0.30 each, replacing $X of headcount."

The contingency line

The budget includes an explicit contingency line, typically 10-15% of total. The contingency exists because the AI ecosystem in 2026 is still volatile — new model releases create migration pressure, security incidents create remediation costs, regulatory changes create compliance costs. A budget without contingency is a budget that will be overrun; the strategist who hides the contingency loses credibility when the overrun happens.

The reconciliation to last year

The budget page references last year's spend and explains the delta. If the budget grew 40% year-over-year, the memo explains why. If it shrunk, the memo explains why. The delta narrative is short — one paragraph — but it is the paragraph the CFO reads most carefully.

The Failure Modes Of Memos

The patterns by which strategy memos fail are predictable. The strategist who watches for them produces a memo executives sign.

The everything-everywhere memo

The memo is twenty pages and lists every agent, every initiative, every consideration. It reads like a status report rather than a strategy. The executive team skims it and remembers nothing. The fix is brutal editing — six pages, the discipline is the point.

The aspirational memo

The memo is written in future tense throughout. "We will build the AI-powered enterprise." "We will lead in agentic intelligence." There are no concrete commitments, no costs, no kill thresholds. The executive team agrees with all of it because none of it requires them to do anything. The fix is to write in committed tense: "we are building X, costing Y, with kill threshold Z."

The technically dense memo

The memo is written for fellow engineers. It uses jargon that the executive team does not parse — "we will migrate to LangGraph 2.x with Helicone-mediated multi-tenant routing." The executive team agrees because they are too embarrassed to ask. They never actually engage. The fix is to write at the level of "what does this do for the business, how much does it cost, what could go wrong" — and put the technical detail in an appendix.

The committee memo

The memo was written by committee. Every stakeholder added a paragraph; nothing was cut. The voice is inconsistent; the priorities contradict each other; the memo says everything and commits to nothing. The fix is single-strategist authorship, with stakeholder review but not stakeholder editing. The memo is signed by multiple people but written by one.

The never-updated memo

The memo was written eighteen months ago and never refreshed. The bets it commits to are no longer the bets the program is making; the budget is from a prior fiscal year; the operating model has changed. The memo is technically still signed but functionally meaningless. The fix is the rhythm page — the memo is updated annually, with explicit refresh date and version number.

The Rhythm Of The Memo

The memo is not a one-time artifact. It is the centerpiece of an ongoing rhythm.

Annual full refresh

Once a year, the strategist writes the full six-page memo from scratch. Not edited from last year's version — rewritten. Last year's memo is reviewed for what aged well and what aged poorly, then archived. The rewrite forces the strategist to re-examine every commitment rather than letting prior commitments compound by inertia.

Quarterly amendments

Every quarter, the strategist publishes an amendment to the memo. The amendment is a one-page document that captures: bets that have completed (with retrospective on outcome), bets that have been killed mid-year (with rationale), new bets that have been added (with budget reallocation), and budget variances (where actual diverged from planned). The amendment is appended to the memo; together they form the living document.

Monthly executive updates

Monthly, the strategist sends a one-page email to the executive sponsors. The email is structured by the memo: progress on each bet, status of each kill, budget vs actual, any new risks or asks. The executives who signed the memo are seeing it operationalized in their inbox every month; the memo stays alive in their minds.

The audit trail

Every version of the memo is preserved with date, signers, and version number. The audit trail matters when the program is under scrutiny — a regulator, an auditor, a new CFO who wants to understand history. The strategist who can produce four years of memos showing the program's evolution has a defensible record; the strategist who has slides only has anecdotes.

Writing The First Memo

The first memo is the hardest. The strategist who has been running the program tactically may have never had to articulate the strategy at the level the memo requires.

Start with the kills

Counterintuitive but consistent: the strategist who starts writing with the kills page produces a better memo than the strategist who starts with the bets. Naming what the program is stopping clarifies what the program is doing; naming what the program is doing without first naming what it is stopping produces a list of additions without focus.

Get the numbers right

The memo's credibility lives or dies on the numbers. Every number in the memo — eval scores, costs per run, headcount, budget — needs to be verifiable. The strategist who puts a number in the memo without being able to defend it loses credibility on the entire memo. If a number is uncertain, the memo should say so explicitly: "estimate based on Q1 data, will refine in quarterly amendment."

Test the memo on a friendly executive

Before circulating the memo for signature, the strategist tests it on one friendly executive — usually the executive sponsor — and asks for honest feedback. The feedback is almost always "too long," "too technical," or "the bets are not specific enough." The strategist edits and re-tests. The first version of the memo is rarely the version that gets signed.

Walk it through, do not just send it

The strategist walks each signer through the memo before asking for signature, in person or on a call. A memo that arrives in an inbox with a request for signature gets less attention than a memo that has been talked through. The walk-through also surfaces concerns before the signature, not after, which prevents the post-signature drift where someone disavows the commitment they nominally signed.

Sign it and circulate broadly

Once signed, the memo is circulated to the program team and to the broader organization with appropriate redactions (confidential budget details may need to be removed for broader circulation). The memo's existence — the program has a written, signed strategy — is the document that anchors every subsequent conversation about the program. New hires read the memo as part of onboarding. New stakeholders read the memo before asking for changes. The memo becomes the program's spine.

The Memo As The Strategist's Defining Skill

The skill of writing the strategy memo is the skill that defines the platform strategist role. A strategist who cannot write the memo can still build agents, can still run platforms, can still consult on operating models. But they cannot get the executive team to commit to a coherent program. The memo is the artifact through which the strategist's understanding of the program becomes the company's commitment to the program.

The hardest part of the role is not the technical sophistication — most strategists arrive with that. The hardest part is the compression: taking everything you know about agents, models, vendors, operating models, risks, and outcomes, and reducing it to six pages an executive will read and one page they will commit to. The compression is the skill. The skill is learnable. The strategist who masters it becomes the indispensable interface between the program and the people who fund it.

Key Takeaways

• The agent strategy memo is a six-page operating contract between the agent program and the executive team that funds it, paired with a one-page board summary.
• The six pages are: state of the program, the bets, the kills, the budget, the risks and asks, the rhythm. Each section is one page; the discipline of fitting is the point.
• The bets-and-kills framing is what makes the memo decisional. Bets are named, costed, kill-thresholded; kills are explicit and signal operating maturity.
• Three to five bets across domain, capability, and frontier categories. Uneven sizing — one big, two medium, one or two small — forces actual prioritization.
• The kills page distinguishes mature memos from immature ones. Four types: agent sunsets, de-prioritized agents, capability investments deferred, program directions closed out.
• The budget is detailed enough for a financial controller, summarized enough for an executive, includes unit economics and a 10-15% contingency, and reconciles to last year with a delta paragraph.
• The one-page board summary has four sections (where we are, what we are doing next, what we are killing, what we need) plus a portfolio table and budget summary visualization.
• Failure modes: everything-everywhere, aspirational, technically dense, committee-written, never-updated. Each has a named fix.
• The rhythm: annual full refresh, quarterly amendments, monthly executive updates, preserved audit trail across years.
• Writing the first memo: start with the kills, get every number right, test on a friendly executive, walk it through with each signer, sign and circulate broadly. The memo becomes the program's spine.