Early AI work optimized for speed. Systems generated output quickly and early signals looked strong. That assumption broke under real usage. When systems misinterpreted intent, overstated certainty, or failed to acknowledge limits, trust eroded and usage dropped.

Working on Agent Name Service pushed that further. Verification solved identity, but not behavior. Systems still failed when language drifted, when responses did not match how customers actually describe their needs, or when uncertainty was hidden behind a confident tone. The trust problem was not about authentication. It was about how the system behaved at the edge of what it knew.

I shifted focus from speed to trust as a system property. Not a feature, not a tone of voice, a structural condition. When identity, language alignment, and behavioral boundaries are explicit, systems hold. When they are not, failure scales faster than any individual fix can address.

The traditional UX model assumed experiences could be validated before release. That assumption failed with AI. Behavior changes based on real usage, not pre-release testing.

I changed the order of work. Designers moved into code across Airo Site Builder, AI Website Builder, and Airo for WordPress — branching and shipping drafts directly in production. I piloted this with leads first, then expanded once behavior proved stable. Feedback loops shifted from review cycles to live signals.

Ownership boundaries flattened. Designers, engineers, and product started operating inside the same system rather than passing work across stages. I moved quality upstream through constraints and shared standards rather than relying on review to catch problems late. The products got better because the team was finally learning from the same conditions the customers were in.

As we scaled across three AI-native builders, the same customer input was producing different outputs depending on which builder interpreted it. The system worked at the surface level but lacked shared understanding of how our customers actually describe what they want when building a website. General language models do not know that vocabulary. The drift was invisible until it compounded.

I unified research, content, and design into one working discipline for the first time. Together we mapped the vocabulary, patterns, and intent signals specific to how our customers describe their goals. That became the Common Language model: a shared reference the AI could draw from consistently across all three builders.

What started as a fix for inconsistent output became infrastructure. Interpretation aligned across builders, AI-generated copy adoption lifted 99%, and the model outlasted the team that built it. Teams outside the original builder group still reference it today.

The first time a designer's undocumented assumption became a live system error, it was clear something had to change. When AI entered workflows that had previously been UX-only, ambiguity functioned differently. Designers could read between the lines in incomplete inputs. AI could not. It filled the gaps confidently and wrong.

What had been minor assumptions humans resolved quietly were now visible system errors with customer impact. I stopped treating context as optional and started building it into the system itself. We built AI companions and shared reference libraries so intent was visible, reusable, and testable.

By 2025 explicit context was no longer a best practice. It was a prerequisite before work began.

After the mobile team moved to my coverage it became clear that the absence of a coherent mobile strategy was causing fragmented decisions and inconsistent execution. The cost was not visible in isolation but it surfaced across platforms and releases.

I helped create a strategy centered on durability rather than perfection. Constraints replaced best practices, guardrails replaced perfection, and decision-making began to stabilize. With VP support the work extended beyond UX and gained broader organizational reach.

Early AI interactions asked users to define intent before the system demonstrated value. This created friction and produced weak signals. Users struggled to articulate what they wanted without seeing what was possible.

I introduced a different approach — "Show, don't ask." Systems demonstrate capability first, then refine based on user response. This reduced effort and improved signal quality. As usage scaled, a deeper issue emerged. Similar language produced different outputs across builders. Systems lacked shared understanding of how customers describe their needs.

I moved language alignment into the system itself. Research, content, and design converged into a shared language model so interpretation became consistent across builders. What started as an interaction pattern became infrastructure. This reduced drift and made behavior more predictable under similar inputs.

UX often influenced direction but did not own measurable impact. That limited its authority in decision making.

I pushed the team to identify work with direct business consequence. One initiative focused on a specific customer experience with clear revenue impact. The team carried it from research through delivery and the result landed above $700K in annual incremental revenue.

This changed how UX operated. Recommendations became inputs to the roadmap and impact became expected. Authority followed measurable outcomes, not participation.

Product teams optimized within boundaries, which created fragmented customer experiences. Each product improved independently but the overall system remained disjointed.

I reframed work around workflows instead of products. Hub unified experiences across surfaces and demonstrated that connected systems produce stronger outcomes than isolated improvements. I applied the same approach across Online Store and Online Appointments, aligning them as one system.

This moved UX from product-level execution to portfolio-level direction, where decisions shape how systems connect rather than how individual features perform.

As the organization scaled globally, execution slowed and quality varied by location. Proximity influenced decisions and ownership.

I addressed structure first by elevating leadership in India to Director level, establishing parity in authority and expectations. We removed regional labels and started treating UX as one system. Directors became accountable for global coverage and shared standards. Product and Engineering aligned through example rather than mandate.

Quality no longer depended on location. It held because expectations were explicit and shared.

A critical initiative moved forward without clear ownership, progress appeared strong but risk was accumulating beneath the surface.

I connected across teams, gathered evidence, and framed the problem in terms leadership could not set aside: customer exposure, product risk, cost of delay. I did not claim ownership. I took it to VP-level leadership, made the risk explicit, and said what others would not. Once visible, ownership snapped into place and the work stabilized before it reached customers.

Systems fail when accountability is unclear. Making risk visible forces alignment faster than assigning responsibility.

Early in my career I believed my intent defined my behavior. Feedback challenged that assumption.

I began separating intent from impact and treating feedback as signal. I examined my language, reactions, and assumptions with the same rigor I applied to design work.

This changed how I lead. I intervene earlier when patterns emerge and adjust based on how behavior is experienced, not how it was intended. That shift improved how I read teams and prevented individual habits from becoming systemic issues.