AI-Enabled Insight & Decision Support
I apply AI as a practical capability inside Agile and program management — reducing cognitive load, accelerating research synthesis, and improving the quality of delivery decisions. My approach is grounded in 25+ years of enterprise experience, hands-on daily use of tools like Claude, Cursor, GitHub Copilot, ChatGPT, and aXet, and a deep respect for governance, process fit, and organizational readiness.
My philosophy
The most dangerous AI implementations I've seen share one thing in common: they were deployed before the decision model was designed.
Organizations rush to adopt AI tools without first clarifying what decisions they're trying to improve, what data they trust, and what governance structure will catch errors. The result is faster noise — not better signal.
My approach starts with the human system: who makes which decisions, with what information, under what constraints. AI is then introduced as a capability that compresses research cycles, surfaces patterns, and structures options — so that experienced leaders can make faster, better-informed calls.
This is not about replacing expertise. It's about amplifying it. The 25+ years of pattern recognition, stakeholder intuition, and cross-sector experience I bring to every engagement is what makes AI output useful rather than just voluminous.
My working toolkit
Reasoning & synthesis
My primary tool for compressing long, messy inputs — stakeholder interviews, retrospectives, PI planning notes — into structured findings. I use it to draft decision memos, RAID logs, and first-pass charters, then apply judgment to what it produces.
Technical discovery
An AI-native IDE I use to read and understand a team's actual codebase — not to write production code myself, but to ask sharper questions in technical discovery. Several of the tools on this site were prototyped here.
Delivery capacity
I don't write code day to day, but I need to understand what Copilot changes about a team's real throughput — helping engineering leads separate a genuine capacity gain from an assumed one before it gets baked into a sprint forecast.
Fast ideation
My go-to for quick first drafts — sprint goal language, retrospective formats, OKR phrasing, scenario questions to stress-test with leadership. Fast and useful for getting unstuck; everything it produces still gets my edit pass.
Enterprise-governed AI
In enterprise engagements with strict data-residency and security requirements, I work inside governed platforms like aXet rather than public tools — a reminder that which AI tool is appropriate is itself a governance decision, not just a preference.
Practical applications
01
Compressing weeks of stakeholder interviews, delivery metrics, and workflow analysis into structured findings — surfacing misalignments across strategy, incentives, structure, and execution with speed that manual methods can't match.
→ Faster diagnosis, sharper problem statements
Tools: Claude, ChatGPT
02
Structuring investment and prioritization decisions by generating, comparing, and stress-testing multiple operating scenarios — so leadership can evaluate tradeoffs with clarity rather than guesswork.
→ Clearer options, more confident choices
Tools: Claude, ChatGPT
03
Identifying structural patterns, workflow bottlenecks, and governance gaps across large, complex organizations — translating raw data into a clear picture of where value is created, where it stalls, and why.
→ Systemic clarity, not anecdote-driven action
Tools: Claude
04
Transforming research findings, delivery metrics, and strategic recommendations into board-ready narratives — structured for executive audiences who need clear problem framing, not data dumps.
→ Faster alignment, better governance decisions
Tools: Claude, ChatGPT
05
Using AI-assisted analysis to model team capacity, dependency risk, and delivery feasibility across large portfolios — giving product and technology leaders a realistic picture before they commit to scope.
→ Commitments grounded in reality, not optimism
Tools: Cursor, GitHub Copilot, Claude
06
Assessing organizational readiness for AI adoption across data quality, governance maturity, process suitability, and leadership alignment — with a pragmatic roadmap that sequences adoption to where it will actually work. Currently serving as AI Agile Delivery Lead on a state government legacy-system modernization program, embedding AI-assisted development practices and executive reporting across multiple workstreams.
→ AI that sticks, not AI that gets abandoned
Tools: aXet, Claude
Prompt engineering in practice
Writing a good prompt and writing a good brief for a team member draw on the same discipline — context, constraints, and a clear picture of what "done" looks like. Program and Agile managers already do this daily; the skill transfers almost directly.
I tell the model who it needs to be and what it's looking at — "You are reviewing a PI planning readout for a VP audience" — the same context I'd give a new team member on day one.
Length, audience, format, and what "done" looks like. A prompt without constraints produces polished-sounding filler; a prompt with them produces something usable.
I feed the model actual retro notes, actual sprint data, actual charter language — never a hypothetical version of the problem. Generic prompts produce generic insight.
The first response is a draft, not a deliverable. I push back, ask for alternatives, and challenge the model's assumptions the same way I'd challenge a team's first pass at a plan.
How it works in practice
Every AI-assisted engagement follows the same disciplined sequence — ensuring that speed doesn't come at the cost of accuracy, and that every output is anchored to the business question it's meant to answer.
AI doesn't make decisions. It makes the humans making decisions better at their jobs — if it's introduced into the right places, at the right time, with the right governance.
— Janet Needham
Where AI gets it wrong
Every one of these tools will occasionally produce a wrong answer with complete confidence — a fabricated statistic, a plausible-sounding dependency that doesn't exist, code that compiles but fails a test it should have caught. In program and delivery work, that confidence is the dangerous part: a wrong number in a status report or a fabricated risk in a steering deck can move real decisions.
I treat every AI output as a draft from a well-read, occasionally overconfident analyst — useful, fast, and never the last check before it reaches a stakeholder.
Fabricated metrics
AI can generate a velocity trend, a completion percentage, or a benchmark that sounds authoritative and isn't. I trace every number back to the source system before it goes in a deck.
Invented dependencies or requirements
Models will occasionally describe a regulatory rule, integration constraint, or stakeholder position that was never actually stated. I verify anything that reads like a fact against a named source or person.
Confident-looking code
Copilot, Cursor, and Claude can produce code that looks correct, compiles, and still fails the case that matters. Generated code gets reviewed and tested like any other contribution — no exceptions for the AI author.
Summary drift
Long synthesis tasks — interview compression, retro roll-ups — can quietly drop context or overweight the most recent input. I spot-check AI summaries against the raw source more than once per engagement.
Guiding principles
I never start with AI. I start with the decision that needs to be made, then determine whether and how AI can improve it.
Every AI use case is assessed for data quality, bias risk, process suitability, and accountability structure before it goes anywhere near a leadership decision.
AI surfaces options and compresses research. A senior practitioner with the right context makes the call. Always. No exceptions.
The measure of AI adoption success is never how much AI is being used. It's whether decision quality improved, cycle times shortened, and business outcomes moved.
AI readiness framework
Stage 1 — Foundation
Establishing data quality standards, process documentation, and decision accountability structures. AI cannot improve decisions built on unreliable data or unclear ownership.
Stage 2 — Augmentation
Introducing AI in specific, well-governed use cases — research synthesis, pattern detection, scenario generation — where human oversight is strong and the decision stakes are measurable.
Stage 3 — Integration
AI becomes part of the standard operating model — integrated into portfolio reviews, capacity planning, and executive briefings. Governance matures alongside capability, and outcomes are continuously measured.
Usually respond within 1 business day