A career that kept increasing in ambiguity.

Building a repeatable way to decide where AI belongs, how much authority it gets, and whether it made the outcome better

I joined to figure out where AI could create real leverage. The evidence changed the question. The constraint was rarely that AI could not do something. It was that the work did not yet have enough context, clarity, ownership, connected systems, or feedback for AI to do it reliably. So the job became understanding how work actually happens, what gets in the way, and what has to be true before AI can improve it.

• 36 structured discovery sessions across GTM, Finance, Marketing, and Engineering in my first few months. I ask people to show me the work, not describe the idealized version of it.
• 23 experiments identified, 4 active pilots by late July. I call them experiments when we are still trying to learn something, and solutions when we already understand the problem.
• Designed a GTM operating system: Claude plus MCPs and skills for reasoning and execution, Notion for persistent context, Slack for human interaction, connected systems for data and actions, and write-back so the record stays current.
• Decomposed GTM into 100+ tasks and 400+ subtasks. "Automate sales" tells me nothing. Task level is where AI decisions become real.
• Built a task-level assessment framework covering understanding, context, AI suitability, integration, notification, autonomy, risk, and value. Capability is not readiness.
• Defined four autonomy levels: human only, human in the loop, human on the loop, fully autonomous. "Keep a human in the loop" is too vague to act on.
• Reverse-engineered ARR and iACV logic from contracts and the Snowflake datamart, turning a week of manual reconciliation into exception review at 89% clean contracts.
• Reviewed the existing AI policy set and asked what kind of thing each document was: principle, norm, guidance, or actual operating rule. Employees need answers at the moment they need them, not a binder.
• Presented the first synthesis to EStaff as "AI Operations: the evidence changed the question," reframing the work as operational readiness rather than a list of use cases.
• Defining the function while doing it: discovery, synthesis, experimentation, prototyping, system design, evaluation, facilitation, enablement, and shared methodology that other people can run without me in the room.

Started with concrete AI problems, ended up building the company's AI operating model

I did not arrive with a theory of AI Operations. I got there by solving enough AI problems to see that the model was rarely the hardest part. Everything around it decided whether it mattered: the knowledge, the workflow, the systems it could reach, who owned the decision, and whether anyone acted on the output.

• Built and ran an AI-first operating model across Product, Engineering, Legal, Marketing, Trust & Safety, Community Operations, monetization, and internal operations.
• Moved the company from scattered individual AI usage to centralized enterprise access and routing. Centralization was not the goal. Visibility, reuse, and cost control were.
• Founded the AI Collective, an open forum for demos, questions, and early feedback, so AI Ops never became the approval desk for experimentation.
• Owned support AI end to end: agent deployment, containment, CSAT, deflection, and whether the customer came back five minutes later anyway. A bot conversation that does not solve the problem is not success.
• Rebuilt support knowledge as infrastructure across a 691-article Help Center: architecture, search, quality, ownership, and the loop from failed answer to new content.
• Connected bot and ticket data in Looker so we could see the whole journey instead of grading the bot and the humans as separate systems.
• Led MAIA, AI-generated content detection: 99.36% accuracy, 99.75% precision, 8 tracks per second, roughly 700K per day, in service of transparency, platform integrity, and differentiated monetization.
• Ran age assurance against UK and Australia regulation, evaluating Persona, Yoti, and VerifyMy against accuracy, privacy, friction, integration, and a first-year cost range of roughly $370K to $860K.
• Shipped AI-supported copyright review, license summarization, and contract indexing so Legal spent its time on judgment instead of retrieval.
• Applied AI to artist distribution and monetization review. Some of the best AI work is invisible: the artist just waits less.
• Designed internal agents and a centralized knowledge model, an early version of the agentic operating systems I design now.

Treated consumer support as a product system across three brands, not a queue

The Weather Channel, Weather Underground, and Storm Radar had consumers moving across brands, platforms, subscriptions, support channels, and help content. I stopped asking how to close tickets faster and started asking why the customer needed a ticket at all, and what that interaction should tell Product.

• Launched an AI-first support entry model with the AI Agent as the primary consumer interface, escalating with context instead of handing humans a blank slate.
• Designed a Copilot model for human agents, because AI should improve the work humans still have to do, not only the work it takes over.
• Defined automation confidence thresholds so "the model produced an answer" never meant "send it to the customer." Answer, assist, escalate, or refuse.
• Wrote AI Operating Procedures covering what the system may do, what source of truth it uses, when a human steps in, and who owns the outcome. That made trust operational.
• Led the redesign of support into a Support Hub combining troubleshooting, product education, value discovery, and expertise, since a customer in support is already trying to understand the product.
• Rebuilt the knowledge loop: product changes, knowledge changes, AI knows, customers get better answers, gaps surface. A stale AI support system is a very confident way to be wrong.
• Established shared taxonomy, metadata, and automated classification so routing, analytics, and product feedback had something reliable to stand on.
• Built the analytics foundation around containment, CSAT, BSAT, deflection, and content gaps, and refused to treat containment alone as proof of success.
• Made support a sensor for Product: if the same issue generates hundreds of cases, the goal is not to get very efficient at answering it forever.
• Got sharp about deterministic versus interpretive work. Exact subscription status does not need creativity. An angry customer with a messy explanation might.

FinTech accelerator; workshops and startup product consulting

At a globally recognized entrepreneurship support organization, developed and led workshops for emerging startups and consulted on product for companies at Little Rock TechPark.

• Designed and facilitated entrepreneurial workshops for early-stage founders.
• Served as product consultant to startups across the FinTech accelerator portfolio.

Multi-country EdTech portfolio, startup through scale

Helped take the company from startup to scale by designing durable product operating models. They kept working, and kept developing people, long after my tenure.

• Drove 400% market share growth: 800 to 4,000 customers.
• Contributed to sustained growth reaching $20M ARR.
• Launched a second major product, scaling from 2 to 436 customers in two years.
• Established product strategy and operating models that became the foundation for long-term growth.

Nonprofit; women's health in the homeless community

Co-founded a nonprofit improving women's lives through direct access to period supplies and women's health resources.

• Built the organization and its direct-distribution model from nothing.
• Focused on women's health access in the homeless community.

E-commerce and mobile in the collegiate bookstore market

Launched and scaled an e-commerce platform generating $1M in first-year revenue, then turned it into a multi-campus enterprise solution. Built the product teams that ran it.

• Initiated the company's expansion into collegiate mobile, earning recognition across the collegiate market.
• $1M first-year e-commerce revenue; scaled into a multi-campus enterprise offering.
• Built and developed high-performing product teams over a decade.