Every quarter, the strategist who runs an agent program has roughly fifteen minutes on a board agenda that is already over-subscribed. The board has questions: How much did we spend on agents? What did we get for it? What blew up? Are we exposed to the regulatory thing the FT wrote about last Tuesday? The temptation is to walk in with twenty-three slides, six product demos, and a slide deck stitched together at midnight that opens with the words "AI is transforming." The discipline is to walk in with three slides — bets, results, risks — plus a standing one-pager on Article 26 readiness, and a clock. This lesson is the three-slide quarterly board update for an agent program in mid-2026: what goes on each slide, what numbers actually convince a fiduciary, what to leave out, how to answer the question the board is going to ask but did not put in writing, and what the standing Article 26 slide looks like when the EU AI Act has been in force for almost a year. The strategist who closes this loop quarterly earns the runway to do the next year of building. The strategist who walks in with vapor loses the program in twelve months.

Why Three Slides and Why This Quarter

A board has somewhere between four and seven hours of meeting time per quarter. The CFO gets the longest slot because the numbers are external. The CRO gets the second-longest because the pipeline is the company. The CTO and CISO trade twenty to thirty minutes between them on platform and security. The Chief Risk Officer or General Counsel takes ten to fifteen on regulation. Everything else fights for the remaining surface area. An agent program is "everything else" until the day it becomes a line item on the income statement, and the strategist's job is to make that day arrive on the strategist's terms, not in the form of a panicked memo from the audit committee after something breaks.

The three-slide format is not a stylistic choice. It is a forcing function. Three slides means three decisions. Bets is the forward look: what we are building and why. Results is the backward look: what we shipped and what it produced. Risks is the lateral look: what could break and what we are doing about it. Anything that does not fit in those three buckets is either irrelevant or belongs in the appendix that the strategist will reference when a board member asks a follow-up. The Article 26 slide is the standing fourth — it is not optional, it does not get a quarterly variance, it gets a status indicator and a one-line update.

The mark of a strategist who has done this before: when a board member asks "what is the biggest risk you are not telling us about?" the answer is on slide three already. The strategist who has not done this before answers from memory and is corrected by the CISO in the next meeting.

What the board actually wants to know

The board is not interested in the agent program in itself. The board is interested in three things about the agent program:

• Is the money producing returns? The CFO will reduce this to dollars in versus dollars out, and so should the strategist.
• Is the company exposed? Regulatory, reputational, customer, security, vendor concentration — the board's fiduciary lens.
• Is the company keeping up? Not "are we innovative" — the board is past that question by mid-2026 — but "are we two years behind the competitor whose announcement I read yesterday."

The three slides answer those three questions in that order. Bets answers "are we keeping up." Results answers "is the money producing." Risks answers "is the company exposed." The Article 26 standing slide pre-empts the regulatory question before a board member asks it.

Slide One: Bets

The bets slide is forward-looking and quarterly-scoped. It names the two to four programs the agent function is actively building or scaling in the current and next quarter, each with a one-line description, an owner, a target outcome with a metric, and a budget. Not seven programs. Not the entire backlog. Two to four programs that the strategist will defend.

Anatomy of a bet line

A well-formed bet line on the board slide reads like this:

• Customer-onboarding agent (production, scaling) — Owner: Priya Ramanathan. Outcome: reduce manual onboarding minutes per new logo from 47 to under 15 by end of Q3. Budget: $340K inference + $180K platform (annualized). Current state: 18 minutes, on track.
• Contract redlining co-pilot (pilot) — Owner: Marcus Chen (Legal Ops). Outcome: redline 80% of NDA and DPA first drafts with a senior attorney accepting the suggestion 70%+ of the time, by end of Q4. Budget: $120K pilot. Current state: 62% acceptance on NDAs, 41% on DPAs; DPA quality work in flight.
• Field-service triage agent (new build) — Owner: TBD (hiring an L4 Staff Agent Engineer; offer out). Outcome: deflect 35% of incoming tickets to self-serve resolution by end of Q1 2027. Budget: $260K all-in. Current state: build kicks off Q3.

Three bets, three owners, three outcomes with numbers, three budgets, three current-state lines. Total slide content fits on a single page in 18-point type. A board member who reads only the slide and never opens the appendix walks out of the meeting able to answer the question "what is the company spending agent money on?" The strategist who answered with vapor — "we are exploring opportunities in customer experience" — gets called back for a second meeting that becomes a re-justification of the entire program.

What does not go on the bets slide

The proof-of-concept that one engineer is building on the weekend does not go on the bets slide. The vendor demo from last week does not go on the bets slide. The internal hackathon does not go on the bets slide. The "we are watching Anthropic's Claude 5 release closely" does not go on the bets slide. The bets slide is what the company is committing money and an owner to. Everything else is research, and the appendix has a one-line research summary if the board has time for it.

The strategist who lists ten bets is signalling that they have not made choices. The board reads the list and concludes the program is unfocused. The strategist who lists two bets is signalling that the program is too small to matter or that the strategist is hiding work. Two to four is the right number for a program in scale-up; one to two if the program is just establishing.

Slide Two: Results

The results slide is backward-looking, quarter-over-quarter or year-over-year. It reports what the bets from previous quarters actually produced. Three numbers matter to the board, in this order: dollar impact (saved or earned), the underlying behavioral metric (the one the strategist tracks), and the cost (inference + platform + people).

Dollar impact is the headline

The board wants the dollar number first. A typical mid-2026 board slide looks like this:

• Customer-onboarding agent — $1.4M annualized run-rate value (240 hours/week of analyst time recaptured, valued at $115/hour fully-loaded, deflected to higher-value work). Q2 2026 actual: $310K against $480K target; on track to exceed target Q3.
• Sales-research agent — $720K annualized run-rate value (35 reps using daily, average 2.1 hours/week recaptured per rep, valued at $185/hour fully-loaded). Q2 2026 actual: $172K.
• Contract redlining co-pilot — $0 yet (pilot). Soft-value: senior attorney NPS for the agent at +47, indicating the pilot is on track for production handoff in Q4.

The numbers are conservative and the methodology is on the appendix. The board is not going to audit the $115/hour fully-loaded rate, but the audit committee will, and the strategist had better have the source — the actual fully-loaded cost from the People team's compensation model — written down.

The behavioral metric is the proof

The dollar number is the headline; the behavioral metric is the proof. For each agent in production, the slide reports the primary behavioral metric the strategist is held to:

• Eval pass rate on the canonical eval set, refreshed monthly. Customer-onboarding agent: 94.3% (vs 91.0% baseline at production launch).
• Severity-1 incidents in the quarter, with a one-line description of the worst one. Q2 2026: one severity-1 (customer-onboarding agent escalated to a human 30 minutes late on May 4; root cause: rate-limit on a downstream tool; fix shipped May 7).
• SLO attainment on the latency and accuracy budgets the AI Council set. Q2 2026: 99.1% latency SLO, 98.7% accuracy SLO, both within budget.

The board does not need to understand what an eval pass rate is in detail. The board needs to see a number that went up, a count of incidents that is honest, and an SLO attainment that is above the budget. The strategist who reports an eval pass rate that dropped, an incident count that is high, or an SLO miss does not hide it. The board respects the honest report and loses respect for the cleaned-up one when the audit committee surfaces the truth a quarter later.

The cost line is the discipline

The cost line on the results slide is the per-agent annualized run-rate. Three components: inference, platform, people.

• Customer-onboarding agent: $340K inference + $180K platform (allocated) + $620K people (2.4 FTE allocated at fully-loaded cost) = $1.14M total against $1.4M value = 1.22x return.
• Sales-research agent: $190K inference + $90K platform + $310K people (1.2 FTE) = $590K total against $720K value = 1.22x return.
• Contract redlining co-pilot: $120K total pilot cost, $0 value yet, expected 2.5x at scale by Q2 2027.

A return of 1.2x is not impressive in isolation, but it is the year-one return on an agent that is still learning and a program that is still maturing. The board reads this and understands the trajectory: production agents are paying back, the program has cost discipline, and the strategist is not selling fairy tales. A board that sees a single agent reporting 8x return without an explanation gets suspicious, and rightly so — the strategist who reports a too-good-to-be-true number is the strategist who is about to be asked to defend it.

The CFO is your ally on this slide if you make them one. Show the methodology before the meeting. Let them push back on the fully-loaded hourly rates. Adopt the changes they ask for. By the time the deck reaches the board, the CFO is nodding through the slide instead of writing notes for a follow-up.

Slide Three: Risks

The risks slide is the one most strategists get wrong. The temptation is to list every imaginable bad thing (cyberattack, model degradation, regulatory action, key-person departure, vendor outage) and rate each red-amber-green. The board reads this and learns nothing. The discipline is to name three to five risks the strategist is actively managing this quarter, with the specific mitigation and the indicator that would tell the strategist the mitigation is failing.

Named risks, named mitigations, named indicators

A well-formed risk line reads like this:

• Vendor concentration (Anthropic). 87% of inference spend on a single provider. Mitigation: dual-route capability through OpenAI gpt-4.1 added to the platform in Q2; one production agent (sales-research) now runs on both, with monthly canary traffic. Indicator-of-failure: if dual-route adoption stalls below 25% of total spend by Q4, the strategist escalates to the CTO. Current state: 6% on the alternate route; on track.
• EU AI Act Article 26 compliance. Customer-onboarding agent operates in employment and education contexts (some customers are universities; the agent makes recommendations that touch student-onboarding flows). Mitigation: human-oversight controls deployed; logging retained for 18 months; impact assessment refreshed monthly. Indicator-of-failure: any deviation in the monthly oversight audit. Current state: zero deviations Q2; standing slide gives full detail.
• Key-person exposure on the agent platform. The platform's tool-calling layer is maintained by a single staff engineer who has been at the company two and a half years. Mitigation: hiring a peer L4 (interview loop active, offer expected Q3); pair-programming on the critical paths; documentation push in Q2 (now 60% complete). Indicator-of-failure: if the offer is not accepted by August 15, the strategist will request a retention bonus from the Comp Committee. Current state: top candidate at final round.
• Litigation exposure on the contract redlining pilot. If the agent surfaces a redline a senior attorney accepts without review and the redline is later challenged, the company's professional liability insurance may not cover an agent-driven recommendation. Mitigation: every accepted redline carries a senior-attorney signature in the system of record; insurance broker engaged Q2 to confirm coverage; pilot scope explicitly excludes regulated documents until coverage is confirmed. Indicator-of-failure: any change in broker guidance. Current state: broker confirmed coverage with no carve-out on June 3.

Four risks, four mitigations, four indicators. Every line is something the strategist could be asked about in two weeks and answer immediately. The board reads this and concludes that the program is being run by an adult. The strategist who lists fourteen color-coded risks with no specific mitigation gets handed a request for a follow-up memo.

The risk the strategist is not naming yet

There is one technique that distinguishes the strategist who has run this loop from the strategist who has not. On the risks slide, name one risk that the strategist is watching but has not yet decided to mitigate. Something like: "We are watching the trajectory of the EU's upcoming general-purpose AI obligations (Article 53 transparency requirements; expected Commission guidance Q4 2026). If guidance requires per-agent training-data disclosures, our customer-onboarding agent is exposed because we use a foundation model whose training corpus we do not control. We are not yet investing in mitigation. If the guidance lands as the worst-case draft, we will need approximately $400K and six months. Decision point: post-guidance, expected November."

This single line does three things. It demonstrates that the strategist is reading the regulatory tea leaves. It pre-empts the board member who is reading the FT and would otherwise raise it as a "have you seen this?" gotcha. It anchors a future budget conversation in advance, so when the strategist comes back in two quarters asking for $400K, the request is not a surprise.

The Standing Article 26 Slide

The EU AI Act has been in force in stages since 2024. Article 26, which covers obligations on deployers of high-risk AI systems, applies to any agent the company operates that falls into the high-risk categories defined in Annex III: employment, education, essential public services, law enforcement, migration, justice, and credit scoring among them. By mid-2026, the question is no longer whether the regulation applies; the question is whether the deployer can demonstrate compliance on demand. The board needs a standing slide that answers "are we compliant?" in under thirty seconds.

What the standing slide contains

The Article 26 standing slide is a one-page status report with seven lines, refreshed quarterly:

• Inventory of high-risk agents in production. "Three high-risk agents in scope as of Q2 2026: customer-onboarding (employment-adjacent for B2B customers in HR-tech), credit-pre-qualification (deployed in two product lines), tutor-recommendation (one education customer)."
• Human-oversight controls deployed. "All three agents route to a named human reviewer on outputs above the confidence threshold. Coverage: 100%. Last audit: June 8. Findings: zero deviations."
• Logging and retention. "All three agents log inputs, outputs, model version, tool calls, and reviewer decisions to the audit warehouse. Retention: 18 months minimum (Article 26 requires six months; we exceed to align with SOC 2). Audit query last executed: June 15."
• Impact assessment refresh status. "All three agents have a fundamental rights impact assessment refreshed within the last 90 days. Customer-onboarding: April 22. Credit-pre-qualification: May 14. Tutor-recommendation: May 30."
• Provider documentation on file. "Anthropic and OpenAI both publish Article 53 GPAI documentation; we have downloaded and archived the current versions. Last archive: June 1."
• Incident reporting status. "Zero reportable incidents in the quarter. The reporting pathway to AI Office through our EU subsidiary is tested annually; last test: March 15, 2026."
• Open compliance items. "One open item: the tutor-recommendation agent's reviewer rotation is currently informal; we are formalizing a written rotation by end of Q3 to remove the audit finding."

Seven lines. A board member who reads only this slide can answer the audit committee's question. A board member who has time can drill into any of the seven lines and ask for evidence — and the strategist who has run this loop has the evidence in the appendix or accessible in the audit warehouse in under a minute.

Why the standing slide is non-negotiable

The first time a regulator, a customer's procurement team, or an auditor asks the company "show us your Article 26 evidence," the strategist who has been keeping the standing slide updated has a ten-minute conversation. The strategist who has not been keeping it updated has a six-week internal scramble during which the regulator's question becomes a finding, the customer's procurement freeze becomes a renewal risk, and the auditor's request becomes a qualified opinion. The standing slide is cheap insurance against an expensive surprise.

The audit committee will eventually ask to see the Article 26 evidence on a meeting's notice. The strategist who has the standing slide in the deck every quarter has been pre-answering this question for a year and walks into the meeting unconcerned. The strategist who only assembled the evidence after the request walks in tense and is asked, reasonably, why the company was not tracking this all along.

How to Rehearse the Board Meeting

The strategist does not walk into the board meeting cold. The pre-read goes out to the board five business days before the meeting, the CFO and CTO and CEO get a dry run two business days before, and the strategist has a stack of three to five anticipated questions with crisp two-sentence answers ready. Rehearsal is the difference between a board meeting that earns trust and one that erodes it.

The anticipated-questions sheet

Three to five anticipated questions, each with a two-sentence answer. The list is shared with the CEO and the CFO so they can prompt the strategist to give the prepared answer if a board member raises the question. Typical mid-2026 anticipated questions:

• "What is our exposure if Anthropic changes its pricing?" Answer: We have dual-route capability to OpenAI on the platform; the cost increase from a 25% Anthropic price hike would be approximately $85K annualized before mitigation, and the dual-route capability lets us shift 40% of spend within 30 days, capping the exposure at $50K.
• "What happens if a competitor announces a new computer-use agent that does what our customer-onboarding agent does?" Answer: Our customer-onboarding agent's moat is the integration with our customers' specific data and workflow; the foundation-model capability would shift faster than our integration depth, so we welcome the capability improvement and would adopt it within a quarter through a model swap.
• "How are we thinking about job displacement among the operations team?" Answer: The 240 hours per week recaptured by the customer-onboarding agent has been redirected to higher-value work, not headcount reduction; we have not laid off a single operations person attributable to agent deployment, and we have re-skilled two ops analysts into our 90-day agent-builder track.
• "Are we training on customer data?" Answer: No. All inference is zero-retention through Anthropic's enterprise tier and OpenAI's enterprise tier; no customer data is used for training. Contractual terms are in the data processing agreements; legal can produce on request.
• "What is the worst thing that has happened with our agents in production?" Answer: One severity-1 incident in Q2 (customer-onboarding agent escalated to a human 30 minutes late on May 4; root cause: rate-limit on a downstream tool; fix shipped May 7; customer was notified and was satisfied with the response).

The pre-read packet

The pre-read includes the three slides, the standing Article 26 slide, the anticipated-questions sheet, and a one-page appendix that backs up every number. The appendix is the strategist's safety net: if a board member asks "how did you derive the $115/hour fully-loaded rate," the answer is on page two. The appendix is short — under five pages — and is built once and updated quarterly, not rebuilt every quarter.

The dry run with the CFO, CTO, and CEO is forty-five minutes two days before the board. The strategist walks through the deck. The executives push on the numbers, the framing, the order of slides. The strategist adjusts. By the time the deck reaches the board, three senior executives have signed off and will back the strategist if a board member pushes hard on something. The strategist who skips the dry run discovers, in the meeting, that the CFO has a different opinion on the methodology, and the board meeting becomes a debate between the strategist and the CFO. This is recoverable once. It is not recoverable twice.

The Quarterly Rhythm

The board meeting is the endpoint of a quarterly rhythm. The strategist who treats the board meeting as the only deliverable will produce a frantic two-week sprint every quarter and will look harried at the meeting. The strategist who runs the quarterly rhythm walks into the meeting with a deck assembled from existing artifacts in two days.

The quarterly cycle from week one to board day

1. Weeks one through ten: the agent program runs its weekly trace review, its monthly eval refresh, its monthly cost-per-run review, and its quarterly Article 26 audit. Each ritual produces an artifact (the trace review summary, the eval refresh diff, the cost-per-agent table, the Article 26 status). The artifacts live in the program's wiki.
2. Week eleven: the strategist drafts the three slides from the existing artifacts. Bets is the program's quarterly roadmap document. Results is the monthly eval and cost artifacts aggregated. Risks is the risk register pruned to the four or five active items. Article 26 standing slide is updated from the latest audit.
3. Week twelve: the strategist circulates the pre-read, runs the dry run with the CFO/CTO/CEO, finalizes the anticipated-questions sheet, and presents to the board.
4. Week thirteen (week one of next quarter): the strategist publishes the board-cycle retrospective internally — what worked, what surprised, what the board pushed on — and updates the quarterly rhythm for the next cycle.

This rhythm produces a board meeting in two days of effort, not two weeks. The artifacts already exist because the program runs its operating rituals continuously. The strategist's job at quarter-end is curation, not creation.

Story: The Quarter the Board Asked a Question

A strategist at a mid-cap fintech walked into a Q1 2026 board meeting with a three-slide deck and a standing Article 26 slide. The deck reported one severity-1 incident: an agent the company had deployed to triage customer complaints had mis-classified a complaint about a fee as a complaint about a marketing email, and the customer had escalated to a regulator before the company noticed. The mis-classification was a single trace, but the customer was angry, the regulator was paying attention, and the strategist had put the incident at the top of slide three with a one-line description, the root cause (a prompt that did not distinguish between fee categories the customer had bundled in a single sentence), and the fix (an added eval case, a prompt update, a confidence-threshold change that routed similar complaints to a human reviewer).

A board member — a former bank-regulator — asked a single question. "If a regulator subpoenas the full audit trail of every customer complaint that agent has touched in the last twelve months, can you produce it in twenty-four hours?" The strategist had prepared for the question and answered: "Yes. The log is in the audit warehouse with eighteen-month retention. The team rehearsed the query in February; the median response time was eleven minutes for a single-customer trace and four hours for a full twelve-month subpoena."

The board member nodded once and the conversation moved on. After the meeting, the CEO called the strategist and said: "That was the question I was hoping someone would ask. You answered it the way I would have wanted to answer it. Keep doing this." The standing Article 26 slide, the rehearsed audit query, and the honest reporting of the severity-1 incident had together earned the program a year of runway. The same incident, reported by a strategist who had not built the rhythm, would have triggered a special audit and a budget freeze.

The board does not punish the strategist for the incident. The board punishes the strategist who hid the incident, or who could not answer the regulator's hypothetical, or who lacked the evidence the audit committee asks for. The three-slide deck plus the standing Article 26 slide plus the quarterly rhythm is the discipline that produces the answer the board wants to hear, not because the answer is rehearsed but because the underlying program is real.

Four Anti-Patterns and How to Avoid Them

The board update goes wrong in predictable ways. The four most common anti-patterns:

Anti-pattern one: the demo deck

The strategist opens a laptop and shows the board a live agent demo. The demo lags, fails to load, or produces a weird answer because the model decided this was the day to hallucinate. The board's attention is now on the failure, not the program. The fix: never demo live in a board meeting. If a demo is essential, record a thirty-second video and play it muted.

Anti-pattern two: the vendor pitch

The slides become a tour of vendors (Anthropic, OpenAI, LangChain, Langfuse, Sierra, n8n, Lindy, Glean, Speak, Brain.fm). The board learns the names of nine companies and the company's strategic position with zero of them. The fix: vendors appear only when they are a risk (concentration, deprecation) or a result (a specific integration that produced a specific outcome). The strategist is not selling for the vendor.

Anti-pattern three: the methodology defense

A board member challenges a number — "where did you get the $115/hour fully-loaded rate?" — and the strategist spends six minutes explaining the People-team compensation model. The conversation never recovers. The fix: the appendix has the methodology in one page; the strategist's answer is "the methodology is on page two of the appendix; happy to walk through it after the meeting if useful." This redirects the conversation to the strategic question.

Anti-pattern four: the hand-wave on Article 26

The strategist says "we are compliant with the AI Act" without showing evidence. The next board meeting, a board member who has been reading regulator commentary asks for specifics, and the strategist scrambles. The fix: the standing slide is in every deck, with the seven specific lines and the appendix backing each one. The board never has to ask because the slide is always there.

This Is the Final Chapter

This lesson is the first lesson of the final chapter of the program. The arc is recognizable. Level 1 was the loop. Level 2 was the eval. Level 3 was the production deployment. Level 4 was the platform. Level 5 has been the strategist's craft: program design, cost discipline, regulatory readiness, talent strategy, and now the rhythm that translates all of that into a board's quarterly read.

The strategist who has built the program is now translating the program for the people whose job it is to be skeptical, fiduciary, and forward-looking. The three slides are the format. The standing Article 26 slide is the receipt. The quarterly rhythm is the operating discipline that makes both possible without burning out the strategist or the team.

The next lesson reads the 2027 roadmap — the three shifts beyond 2026 baseline of A2A and durable execution that the strategist is starting to plan for now. The lesson after that is the strategist's external voice: the talk, the post, the open-source contribution that signals to the broader community what the strategist has built and what the strategist believes. Then the program ends, and the strategist returns to their company with a year-long runway, a written program, and a board that trusts them.

Key Takeaways

• The board has fifteen minutes for the agent program. Three slides — bets, results, risks — plus a standing Article 26 slide is the format. Anything else is appendix.
• The bets slide names two to four programs, each with an owner, an outcome with a metric, a budget, and a current-state line. Two to four, not ten.
• The results slide leads with the dollar number, backs it with a behavioral metric (eval pass rate, incident count, SLO attainment), and reports the all-in cost per agent.
• The risks slide names three to five risks the strategist is actively managing, each with a specific mitigation and an indicator that would tell the strategist the mitigation is failing.
• The standing Article 26 slide is seven lines: inventory, oversight, logging, impact assessment, provider documentation, incident reporting, open items. Refreshed quarterly.
• The board respects honest reporting of the bad number and loses respect for the cleaned-up version when the audit committee surfaces the truth a quarter later.
• The dry run with the CFO, CTO, and CEO two days before the board meeting is non-negotiable. Three executives sign off before the deck reaches the board.
• The quarterly rhythm produces a board meeting in two days of effort because the operating rituals (weekly trace review, monthly eval refresh, monthly cost review, quarterly Article 26 audit) generate the artifacts continuously.
• The board punishes the strategist who hides the incident, who cannot answer the regulator's hypothetical, or who lacks audit evidence. The board does not punish the strategist who reports honestly and is prepared.
• The strategist who runs this loop quarterly earns the runway for the next year. The strategist who walks in with vapor loses the program in twelve months.