TL;DR
One-line outcome. Rebuilt the product foundation in 90 days, then doubled AI usage across the platform.
Role, dates, scope. Sr. Director of Product & Design at Axios HQ, March 2023 to June 2025. Led product, design, research, and analytics. Reported to company leadership. Worked directly with the CEO, CMO, and Head of AI on product direction and AI strategy.
Metric chips.
- Doubled AI usage across the platform
- Rebuilt the product information architecture in 90 days
Context
Axios HQ had outgrown its original shape. What started as a focused writing tool had become a platform in waiting, and every new feature was fighting the old structure of the app. The team was shipping. The product was getting harder to extend with every release.
The real problem was not navigation. It was that the product no longer had a mental model anyone could hold in their head, and the company was about to ask it to carry a much bigger roadmap.
Mandate
Give the company a structure it could build on, set the operating model around it, and make AI a real part of how the product worked instead of a press release. The job was not to ship more features. It was to give the product a backbone it did not yet have, and to build the team that could use it.
Scope
I led product, design, research, and analytics as one organization, not four workstreams. Each function was built and matured in place, not run in parallel and stitched together later.
I reported to company leadership and worked directly with the CEO on product direction, the CMO and Head of AI on AI strategy, and engineering leadership on platform decisions. Authority covered what we built, how it was structured, and how the team made decisions together.
Key decisions
Rebuild the information architecture before adding features. The product had stacked complexity for years and the pull was always toward shipping more. I stopped and named the structural problem first. The IA got rebuilt in the first three months so the AI work that came after had a coherent shape to live inside, not bolted on top of an app that no longer made sense.
Visible and invisible AI as the strategic frame. AI needed a lens that could guide individual product decisions without me or the CMO arbitrating each one. I shaped a "visible versus invisible AI" frame with the CMO and the Head of AI. AI that earns user trust surfaces itself. AI that earns user time stays out of the way. That frame guided the product surfaces and the internal automation behind them.
Shape Up and North Star as planning anchors. The team needed a planning cadence it could own. I introduced Shape Up for product cycles and North Star framing for strategic alignment, so the team could commit to bets without losing visibility for executives.
Research as infrastructure, not a project. The research function got built to produce durable assets the team could draw on continuously: personas, JTBD frameworks, decision artifacts. Not one-off studies that get archived and forgotten the week after the readout.
Before / 2023
One root. Nine leaves. Constant collisions.
Every new feature hung off the same overloaded root. Cross-links between siblings made each addition a structural decision.
After / 90 days
Three pillars. AI at the right tier.
Plan, Compose, Send and Measure. AI threaded into Compose where users see it, and into Send and Measure where it works ambiently.
What changed
- The product information architecture got rebuilt. The team had one map to work from, and the one-off structural decisions every new feature was forcing came to an end.
- Research function built from scratch. Personas and JTBD work grounded the team in evidence, not opinion.
- A planning cadence the team could own. Shape Up shaped product cycles. North Star framing kept executives informed without forcing the team to defend the work every week.
- AI strategy held a clear frame. Visible AI shipped where it earned its place. Invisible AI shipped where it removed friction users never had to think about.
- Internal AI adoption pushed hard. The team felt the same productivity lift we were trying to ship to customers.
v1 · 2022
v2 · 2024
Selected v3 surfaces
What the rebuild made possible.
- 01 / 03Vision Planner
Editorial planning by week and channel. The kanban surfaces commitment, not just intent. - 02 / 03Calendar
Schedule-send and edit-target-date in the same context menu. The calendar reads as a budget, not an inbox. - 03 / 03Collaboration
Threaded inline comments on the highlight, with mentions that route the question to the right teammate.
Measurable outcomes
AI usage across the platform
Doubled inside one tenure.
Verified from source material.
- Doubled AI usage across the platform.
- Rebuilt the product information architecture inside the first 90 days.
- The product matured into something the company could fund and scale, with a real research function, a real planning cadence, and an AI story that held up under investor questions.
Leadership lens
The fastest way to slow a product down is to stack features on a structure that no longer fits. The instinct at Axios HQ would have been to ship harder into a product that was already congested. I named the architecture as the constraint and spent the first three months solving for it. The AI work, the research build, and the planning model all sat on top of that decision. Make the structure visible before you make the output faster. That sequencing is the leadership move, not the architecture itself.
What I did with my hands
Player-coach proof. The work was authored, not just sponsored. I drafted the IA rebuild, wrote the visible-versus-invisible AI frame with the CMO and Head of AI, ran the JTBD research the team operated on, and held the merge button on the planning artifacts. Nothing in this case study lived in a slide deck I asked someone else to build.
AI threading
The "visible versus invisible AI" frame is the strategic artifact of this work. It is what made the doubled-usage outcome possible: a durable principle the team could apply without an executive in the room. Visible AI shows the user that intelligence is in the loop, so trust gets earned consciously. Invisible AI removes friction without asking for credit, so user time gets returned.
The frame shaped two streams of work. Product surfaces where AI assistance was visible to end users. An internal operating layer where Zapier flows and custom GPTs absorbed the manual overhead inside the team. Both streams shipped against the same rule.
The Strategic Frame
Visible vs. invisible AI
A two-axis decision rule for every AI surface and internal flow at Axios HQ. Show the work where the user owns the decision. Disappear where the user wants the friction gone.
Visible AI
Earns trust by surfacing the intelligence.
Where AI participates in a decision the user owns, the system shows its work and gives the user the controls to accept, edit, or reject.
01 Smart Brevity panel
Subject-line suggestions with Dismiss and Regenerate.
02 ROI dashboard actions
AI-suggested next steps such as "Send the Audience Survey."
03 Insights and recommendations
Surfaced under a sparkle icon so the user sees the assist.
Invisible AI
Earns time by removing the friction.
Where AI removes a step the user never wanted to take, the system performs the work in the background and leaves the override visible but quiet.
01 Editorial planning automation
Cadence work the editor never has to assemble by hand.
02 Suggested chips on metadata
Pre-filled tags in the editor metadata column.
03 Internal Zapier and custom GPTs
Background flows that absorb the team's operational overhead.
Reflection
I would sequence the research build before the IA rebuild, not after. The JTBD assets and personas that came out of the research function validated decisions we had already made on gut. Starting with user mental models would have sharpened the architectural choices from day one and cut some iteration downstream.