Arushi Tandon
AI-first product designer
AI-native designer shaping the future of spreadsheeting on mobile and iPad.
Senior Product Designer at Microsoft. I lead design for Excel on mobile and iPad, from vision to launch.
A1
Arushi Tandon
AI-first product designer
Senior Product Designer at Microsoft. I lead design for Excel on mobile and iPad, from vision to launch.
Senior Product Designer at Microsoft, five years and counting. I lead design for Excel on mobile and iPad, from vision to launch. AI-first product designer: I design in code and in systems, I build the AI tools my team uses, and I lead a team of designers who now fix craft bugs directly in code. Gurgaon, India.
View my CV · [email protected] · LinkedIn
Spreadsheets on mobile were built for reading, not doing. I set the design vision for Excel Agent on iPhone and iPad and led it from research to launch. People trust an agent by checking its work, so every surface makes the agent's state legible at a moment of decision: clarifying questions before it acts, steps while it works, change highlights with Done and Undo after, and fire-and-return for people who switch apps.
My role: a triad effort with product and engineering. I led design with my team of three designers: I set the strategy, did the systems thinking, and drove the collaboration and alignment updates across partners.
Key decisions: trust scales with risk (sort and filter run straight away, formatting shows a preview, formulas and multi-step work go step by step); modes as a scaffold (Edit, Plan and Chat ship for reliability, with clarifying questions built underneath so the modes can recede); design for both waiting styles (the Monitor who watches every step and the Backgrounder who returns on a notification).
Outcomes: live on iPhone and iPad; growing organically, with numbers shared in conversation; research with 200+ users surveyed and 15 moderated sessions. Honest limit: agent latency still sits above a typical mobile session; design stages it and covers it with notifications.
I moved my team from designing in tools to designing, auditing and fixing directly in code. I built HAWK AI, an internal Microsoft tool that watches recordings of Office apps and files the UX defects it finds with visual evidence and severity. I built it in September 2025; once it was funded in February 2026, engineers scaled it to production, and I still lead its design and own it. PRISM is where designers design directly in code and hand off a validated package to engineering. The Microanalyzer audits a few seconds of an interaction frame by frame. My design team now fixes every craft bug in code across the Excel app, the previewer and Excel Agent.
Key decisions: define "good" before scaling (thirteen binary checks weighted into bug-finding accuracy 40%, triagability 30%, hallucination control 30%); document by-design behaviour as context rather than engineer it away; crawl, walk, run, scale toward Word and PowerPoint.
Outcomes: 4–5 days saved per audit cycle; 7,000+ videos analysed; 25+ critical issues caught in the first run; 40+ bugs fixed. Honest limit: accuracy on the previewer sits around half, and I shared that in the leadership update.
Copilot could draw a chart nobody could explore. I designed the interactive chart experience in Copilot chat and ran the research behind it: about ten interviews across the US, the UK and Asia. Four hypotheses, four overturned: interactivity is a basic need; Excel needs its own definition of interactivity because it serves makers and readers on one surface; consumption dictates creation; and people do light reporting in Excel. The feature didn't ship because the team moved to the agent, which can change the grid directly. The research findings still hold.
Cell after cell of numbers does not tell a story. At a Microsoft hackathon I reimagined Excel as a Copilot-first data hub: one prompt becomes a visible plan, each step previews its result and asks before changing the data, and the results become a report with interactive charts that the reader can question in place, shared as a workbook, a deck, a document or a PDF. Patent application filed May 2025.
Every app an organisation runs has a gatekeeper. I owned nine features end to end in the Microsoft 365 admin centre and ran a usability study with admins at six global enterprises: they navigated by search while app state lived in tabs. Two of my three recommendations shipped: status in the available apps list, and deploying from the app overview. Unblocking restores an app to its last saved state, so undo comes without loss. In 2023 I designed the governance layer for AI agents: a tenant default, a pilot group, then per-agent review of capabilities, knowledge and actions before publishing. My own popular-apps recommendations added no value, because enterprise adoption runs through approvals; the study showed it and I said so.
I didn't just adopt AI. I changed how I design. Designer, then designer with AI. Now, AI-first. This site is designed and built in code.