Enterprise software, rebuilt around agents.

The bet

Enterprise software is being rebuilt around agents — human and AI, working side by side. The advantage goes to companies whose platforms were architected for that partnership from the start, where people and AI amplify each other's work, rather than the ones retrofitting AI onto what they already had. My role was less about designing a thing and more about understanding what every business unit needed, then pulling those needs back to the customer's lens until dozens of teams converged on one coherent experience — built for the people using the platform, not the org chart that produced it.

The before state

ServiceNow's platform UI accumulated over a decade, for good reasons. Core UI (2013) was the central place customers got work done — dashboards, lists, forms. It was a simple, consistent system, and that consistency is what drove its broad adoption. Over time, the platform added admin tools, builder tools, and configuration interfaces. By 2020–2022, the shift came with workspaces: business units needed market-specific, tailored experiences, so workspaces became configurable, extensible, and BU-owned.

But that flexibility had a cost. Two things became visible: proliferation (70+ workspaces, most of them fulfiller-facing, spread across many BU areas) and usage drift (customers outsourced workspace development to MSPs because configuring them required real expertise).

The diagnosis

Before any of this could be argued, the organization needed a shared, defensible picture of the problem — and it didn't have one. No canonical inventory of the platform's surfaces existed. Estimates of how many workspaces had shipped ranged from "a handful" to "a lot," and for a meaningful share of them, finding the current owner meant tracking people down one by one.

I initiated and directed the diagnostic; my team built it, partnering with BU designers across the product areas. We created what we called the Experience Library — a canonical catalog of the platform's experiences, assembled from screenshots of live customer instances, not Figma mocks. Capturing real pages from real environments let us put canonical patterns and their outliers side by side, and it exposed how often teams were piecing components together by hand instead of using the native ones. It surfaced 70+ workspaces across more than twenty product areas and became the source of truth the platform had never had.

Alongside the new audit, we pulled years of prior cross-BU research — agent segmentation, workspace-adoption studies, interoperability surveys — into a single evidence base. The audit got the attention, but the library was the lasting asset — and pulling the prior research in alongside it is what moved the conversation from opinion to evidence.

The three hard problems

One — The surface people touched

If the platform became genuinely agentic, what did the user's home base look like? The old answer — navigate to the workspace you need; the workflow begins at the record level — was a workflow-era artifact. The new answer inverted the hierarchy: workflows surface up from the record level into a new home that sits above all workspace archetypes.

Two — The governance patterns that made it trustworthy

Enterprise customers will not deploy agentic software they cannot audit. Two ends of the spectrum already existed: fully deterministic workflows driven by playbooks, and purely human, manual workflows. The middle — where the system proposes and a person reviews, approves, or overrides — was the part with no shared design vocabulary, and it was where the new work lived.

My team worked the experience and design side of it. We designed the interaction patterns for those human-and-AI workflows — how a recommendation carries its reasoning, how a person reviews and acts on it, how an override is captured and made visible — and contributed them into the Experience Library so any team building on the platform could adopt them. Explainability was a design primitive in these patterns, not an add-on. The deterministic and purely-human ends were already solved; the design contribution was the part in between.

Three — The operating model that produced it

40+ scrum teams don't converge on a coherent platform experience by accident. The old model — component-based release planning, each team shipping its own pieces — produced exactly the fragmentation we'd diagnosed. I drove the shift to outcome-aligned workstreams: teams organized around the experiences customers actually had, not around the components that happened to compose them. In practice it ran as three parallel workstreams — shipping AI Control Tower as the first product on the new platform layer, consolidating the fulfiller surfaces, and completing the re-platform onto one shared foundation. The reorganization is what made the rest shippable.

How the surface was designed

The proactive-surface concept needed concrete principles to design against. Four held across every surface AINPX shaped:

• Outcome-centric. Surface what the persona needs to do next, not just everything the platform knows.
• Progressive disclosure. Start with the essential; let the user drill deeper on demand. The agentic surface is dense; the rendered surface is restrained.
• AI and agents as accelerators. Chat, suggestions, and automation reduce friction at the moments where it was costing time — load-bearing AI, not decorative AI.
• Continue where you left off. The home helps someone pick up work mid-stream rather than start over. The platform respects state and context.

The home itself was built around four ranked modules — Today's priorities (the default surface), Quick wins (low-effort tasks that build momentum), Pending tasks (follow-up work to return to), and Clustered insights (affinity-grouped work the user can clear in a single context). Each carries an explanation of why an item is there and what to do with it. That is what "the workflow surfaces to the user" looks like once it's designed — a specific home rather than an abstraction.

Rebuilding design for the AI era

Coherence at the surface means nothing if the practice behind it still moves at the old speed. The traditional double diamond — discover, define, develop, deliver — assumes a linear path from problem to shipped product, and that assumption is under real pressure: engineers prototype in hours, designers spin up ten directions in the time it once took to build one. The proportions of the work shifted — less time mocking, more time giving direction and guiding quality. So we rebuilt how design got made.

Handoffs were the first thing to go. Product, engineering, and design now work as a single triad with shared accountability for every decision — no throwing work over a wall. The rhythm is tight by design: builds run overnight, the triad reviews each morning, a short sync aligns on changes, and the next build is queued by end of day — with design leading the quality bar inside that partnership. How the triad works became as much the deliverable as what it built, and that working model is now the standard for how platform component work gets done.

The build workflow itself became a contribution loop. Every team starts from the same question: does the shared design system already provide what this experience needs? If it does, they use the building blocks — and where no native pattern exists, AI generates a compliant variant rather than a hand-built one-off. When a pattern proves common across business units, it's standardized and contributed back into the shared system, so the system compounds instead of bottlenecking on a central team. The aim was a design practice that keeps pace with the rest of the product.

What changed

A few shifts outlasted any single shipped feature:

• AINPX became the canonical AI platform layer at ServiceNow. New AI capability now lands in one governed, coherent layer instead of scattering into BU-specific pockets — a decision about where the platform's AI future gets built. The wedge that first earned the mandate was an early interoperability win: a modest stopgap in the old multi-workspace model that proved the consolidation case.
• AI Control Tower was the first product to ship on it — unveiled publicly at ServiceNow's Financial Analyst Day, the point where the architectural promise turned from internal aspiration into a public commercial story.
• EmployeeWorks shipped within it to 350K frontline employees at a major Fortune 500 deployment, seeing 50+ opens per shift — turning mobile from a compliance checkbox into the primary surface for frontline work.
• A validated proof-of-concept showed 10× productivity gains on real enterprise workflows — evidence the operating-layer thesis was defensible, not just aspirational.

Why this matters

The bet paid off in public. Internally, the founding principle was blunt: AI is the operating layer, not a sidecar bolted onto each product. At Knowledge 2026, ServiceNow said the same thing from the keynote stage — positioning itself as the "AI control tower for business reinvention," with a single unified experience layer across every workflow and AI built in natively rather than added on. The thesis AINPX was built on — one coherent, governed surface where the work surfaces to the person — had become company strategy. What started as an internal initiative became a company-level priority, and the products it seeded are central to how ServiceNow now describes its future.