Director, AI Strategy & Governance · Enterprise AI · GenAI · AI Transformation
AI strategy and technology executive with 25+ years across enterprise software, AI/ML, business intelligence, e-commerce, and business ownership. Advises executives, boards, and technical teams on AI strategy, enterprise architecture, GenAI adoption, AI governance, and technology-enabled transformation across healthcare, insurance, government, and Fortune 500 environments. Dual MBA (Columbia, Finance & Economics; UC Berkeley Haas, Strategy & Entrepreneurship) and a computer science degree (UC Davis) underneath the strategy work, not instead of it.
A few of the numbers, plainly stated:
- Architected a municipal multi-agent AI system that reduced help-desk workload by 23%, after two prior enterprise vendors had already failed at it.
- Designed an enterprise RAG knowledge-retrieval pipeline that improved decision accuracy by 85% against a four-year baseline.
- Diagnosed a clinical ML accuracy gap to its real root cause (data architecture, a sharding scheme, not the model) and redesigned the pipeline, taking prediction accuracy from 88% to 98% at Change Healthcare, with $3M+ in separately identified savings.
- Directed a ServiceNow AI workflow redesign that increased first-touch resolution by 65% and generated $2M+ in documented labor savings.
- Built predictive AI models for real-estate investment analysis that surfaced $50M+ in acquisition opportunities.
- Co-founded an online music company (massmusic.com), scaled it to a Forbes #3 ranking in online music sales, and negotiated its acquisition.
Currently: independent AI strategy consultant, and co-founder and fractional CTO of 150LEFT, an AI-powered insurance risk-assessment platform (predictive models improved risk prediction accuracy by more than 50%; GenAI agents on AutoGen, MemGPT, and CrewAI). Earlier: Unify Consulting (GenAI strategy for Fortune 500 clients), engineering leadership at AudioHighway.com, Grapevine, and Art.com, and a start in anti-virus research at McAfee. Full history at linkedin.com/in/scottcole.
Author of 31 books, centered on the five-book Stop Learning AI series, written for executives who need to make good AI decisions without becoming technical themselves:
Also on AI strategy and adoption:
As an Amazon Associate I earn from qualifying purchases.
Not a developer portfolio; the opposite problem, actually. Vendor evaluation and governance advice is only as good as whether the advisor can tell a real capability from a demo. These are how I stay able to tell the difference myself, not delegate that judgment to whoever's selling.
| Project | What it actually is |
|---|---|
| Evo | LLM-driven tribes grow from a handful of survivors into a founded, warring, or allying civilization, entirely on their own. Seven-era progression, diplomacy that can genuinely backfire, a headless benchmark harness for reproducible model-vs-model comparison. |
| Void-Marauders | A sci-fi colony sim where autonomous agents explore, build, and can mutiny. Loyalty is outcome-sensitive, not a flat scoreboard: a risky order that gets someone hurt costs a captain real, measurable trust. |
| Palimpsest | Durable, curated memory for AI agents. Every new claim is judged against what the memory already holds (reinforce, collide, or hold for a person's review), so a forged fact can't quietly rewrite it; an optional model check may raise a flag but never clear one. Tested against one falsifiable bar: does having this memory change a judgment, not just "does storage work." Latest: v0.3.0. |
| Aegis Vector | Plants forged policy documents in a local RAG pipeline and measures, defense by defense, how often small models believe them and what stopping them costs per query. Across nine forgeries, the one defense that got adoption to zero without refusing answers was Palimpsest, used as a memory gate; a forgery that states no number at all was still adopted under every other defense tested. Includes a live dashboard and an animated Black Hat vs White Hat episode player built from the real runs, where every loss sends White Hat up a ladder of different defenses until one holds. Latest: v0.2.0. |
| Executive Decision Intelligence | A local decision advisor for small-business capital decisions: forward-looking vs total NPV to break the sunk-cost trap, Monte Carlo risk, a synthetic customer focus group, and a red-team challenger that argues against the pitch. Its cases include one where the numbers say avoid despite a hard vendor pitch, because an advisor that only ever says yes isn't one. |
| life-rolls | Five or more distinct local models bluff and call each other out in a live Liar's Dice tournament. Zero external dependencies: arithmetic always happens in real code, never asked of a model. |
| Dominion | Thirteen contestants, each backed by a local model, draft trivia domains and duel for a game-show grand prize, with a scripted fallback so the show never stalls if a model call fails. |
Credentials without hands-on depth get talked past by a technical team. Depth without the credentials gets talked past by the board. Both are real here, on purpose.