Miku Jha

Continuing to noodle on the Personal AI assistant market and its many open questions: Will we see see one assistant per person? Two plausible architectures: - AI Chief of Staff: A persistent entity maintains context across your life. - The agent org: Specialized agents for travel, finance, shopping, health, coordinated by a meta-agent. Google is already pointing toward customizable sub-agents inside Spark. My guess is consumers experience one assistant, with specialized systems underneath. What’s the eventual business model? - Subscription: “Pay $50–200/month to save hours.” - Transaction fees: Book a hotel → AI gets 5%. Buy shoes → affiliate revenue. - Advertising: “Here’s the restaurant I think you’d like.” Is it the best choice for you, or the best monetized choice? Search can tolerate conflicts because users know it's a marketplace. Personal assistants implicitly promise they’re acting in our interest. That makes alignment a differentiator. Will assistants become buyers rather than recommenders? Today commerce is optimized for humans: SEO gets the click, ads persuade, landing pages convert. But if I tell my assistant: “Find me a hotel in London near Mayfair, <$800, quiet, good gym,” I may never enter the sales funnel. That changes the optimization target for merchants: from persuading humans to winning over their agents. I would distill the competition down to 6 assets: 1. Context: does it understand me? 2. Agency: can it accomplish things? 3. Reliability: can it to do the right thing consistently? 4. Trust: am i comfortable giving it authority? 5. Distribution: does it exist where i already live? 6. Taste: can it make the right choice for me, when there is no right answer? The winning formula is multiplicative. If any term approaches 0, the product collapses. Extraordinary intelligence but no permissions - chatbot Extraordinary permissions but no reliability - dangerous Great reliability but no personal context - automation tool Deep context but no trust - creepy Deep context but no taste - personalized mediocrity Great taste but no agency - concierge without hands Everything except usage frequency - demo Questions I’d ask when under-writing this category: - Does retention increase as the assistant learns the user? That tells you whether a real context/memory flywheel exists. - What % of tasks complete without intervention, and how quickly does that improve? That’s the agent equivalent of model quality. - How far up the trust ladder do users move? Reading email is one thing; sending email, spending money are stronger signals. - Does usage broaden over time? A great assistant should land with 1-2 use cases and absorb more of your life. - Can a startup accumulate enough proprietary context/relationship advantage before incumbents close the gap? The existential question I think we are seeing the beginnings of a new consumer operating layer. Unclear whether it belongs to model providers, operating systems, or independent assistants.

Is System Design Becoming a Lost Art? I've been noticing a view gaining traction in interviews and on this platform: that system design is a relic. The reasoning goes something like — "AI can architect this for me now, so why practice it?" I think that's a dangerous misreading of what we're actually valued for. Why "AI will design it" doesn't hold up In the enterprise world, we rarely work with clean requirements. We deal with: - Legacy security postures that rule out the "ideal" tool - Constraints that demand creative workarounds, not textbook solutions - Unstated requirements that no prompt will ever surface AI is genuinely excellent at pattern-matching to a reasonable architecture. What it can't do is anticipate the specific failure modes of your business logic, or judge whether a design will still be maintainable in three years under your team's constraints. That judgment is the job. The real risk isn't AI — it's atrophy If we stop practicing system design because a tool can produce something that runs, we don't just lose a skill. We lose the thing that made us valuable to the enterprise in the first place: the ability to catch bad logic before it ships, not after. Where I land Don't outsource your judgment. Use AI to move faster — never as a substitute for thinking through why a design works. Fundamentals still compound. Structured system-design thinking (frameworks like Lewis Lin's are a good starting point) isn't just interview prep — it's the muscle you're actually being paid to have. The real test isn't "does it run." It's whether the result is maintainable, scalable, and secure within your specific constraints. That's still a human call. The bar isn't lowering in the age of AI — it's rising. The architects who matter most going forward won't be the ones who can prompt a design into existence. They'll be the ones who can audit it.

I am ecstatic to announce that I was selected to participate in the inaugural Gap Doris Fisher Creators Program cohort. I am so excited to learn from expert designers at the Gap and experience firsthand what makes it such a successful company. Though we are only in the early stages, I can tell the program will be an invaluable experience and will provide crucial feedback on my work!

An always-on AI workforce meets the city that never sleeps! It was great to be back in New York for ServiceNow World Forum, where I presented our spotlight session, “Data & Analytics: Agent-ready data for the Autonomous Workforce.” For AI agents to act with confidence, they need the same context your best people rely on: customer history, company policies, approval paths, and past decisions. Without that understanding, even the most impressive reasoning can amount to little more than confident guessing. Getting your enterprise data agent-ready — and keeping it that way — means connecting it, governing it, giving it meaning, and bringing insight closer to action. That’s why we’re uniting data, analytics, and workflows on the ServiceNow AI Platform, so people 𝘢𝘯𝘥 AI agents can make informed decisions and act on them, right where work happens. Huge thanks to Tim Gasper for leading our fireside chat, and to Mal Vivek, CEO of zeb, for joining us and sharing how her team is putting ServiceNow to work. World Forum season is one of my favorite times of the year. What I value most is connecting with customers and partners, hearing what they want to achieve, and exploring how we can help. In New York, we also hosted our annual Global Partner Ecosystem Summit, led by Michael Park. I came away energized by a day of candid conversations about how to move fast with AI and deliver more for our customers while maintaining trust. Thank you to everyone who joined us, and to our entire team for the work behind every story we get to tell. Next stop: the inaugural Mumbai World Forum on October 6, where I am looking forward to delivering the keynote along with Amit Zavery. I hope to see you there! Find out more: https://lnkd.in/grqKACYS #ServiceNowWorldForum Michael Meiner Lisa Berlin Bipin Singh Alex Elliott Pavithra Pandian, MBA Melody Mileski Gina Mastantuono Steve Walters John Aisien Jeffrey Hausman Kellie Romack Jon Sigler Nenshad Bardoliwalla

Absolutely humbled to have received the 𝗕𝗲𝘀𝘁 𝗣𝗮𝗽𝗲𝗿 𝗔𝘄𝗮𝗿𝗱 🏆 at the IEEE International Conference on Image Processing (ICIP) 2026 for our AI research work. It's a surreal feeling to have won this award competing alongside some of the top AI talent globally, from various top universities and companies across the US, Europe, and Asia. Abhijat and I presented our paper on 𝗙𝗿é𝗰𝗵𝗲𝘁 𝗪𝗮𝘃𝗲𝗹𝗲𝘁 𝗦𝘁𝘆𝗹𝗲 𝗗𝗶𝘀𝘁𝗮𝗻𝗰𝗲 (𝗙𝗪𝗦𝗗): 𝗔𝗻 𝗜𝗻𝘁𝗲𝗿𝗽𝗿𝗲𝘁𝗮𝗯𝗹𝗲 𝗜𝗺𝗮𝗴𝗲 𝗦𝘁𝘆𝗹𝗲 𝗦𝗶𝗺𝗶𝗹𝗮𝗿𝗶𝘁𝘆 𝗠𝗲𝘁𝗿𝗶𝗰.You can read more about our work here: https://lnkd.in/dXgJaEg3 Proud to have co-authored and presented this work with Abhijat Bharadwaj, and immensely grateful to our co-authors #AnimeshKumar and #ProfVikramGadre for their guidance and collaboration. The award was presented during one of the most regal gala dinners I've ever attended an unforgettable setting for an already unforgettable moment. Also had the incredible opportunity to meet #ProfStéphaneMallat, whose pioneering work revolutionized the foundations of multiresolution analysis, the very field that I've had the privilege of bringing together with Generative AI throughout my PhD. A heartfelt thank you to Praveen Damodaran, Miku Jha and everyone who supported us throughout this journey. 🙂 #ICIP2026 #IEEE #AIResearch #ImageProcessing Praveen Damodaran Miku Jha John Aisien Joe Kotler

When enterprises evaluate Voice AI, one of the first questions is rarely, "Can it hold a conversation?" It's "Will it work with my contact center?" Our approach is simple: meet customers where they are and integrate with the telephony and contact center platforms they already trust. In this blog, I discuss how we are helping customers bring AI Voice Agents into their existing telephony ecosystem, from leading CCaaS platforms to regional providers and customer-managed environments. With Custom SIP, available this September, customers gain even greater flexibility to connect ServiceNow Voice AI to bespoke architectures and long-tail telephony providers. Learn more: https://lnkd.in/g8S52cxX #VoiceAI

AI Value gap is not always technical ! Most people assume Forward Deployed Engineering (#FDE) means walking in and solving a problem the customer already defined. In the enterprise, it's usually the opposite. Last month, I opened an engagement with a large, multinational enterprise — no problem statement, no use-case shortlist, nothing to react to. Just a live platform and a lot of ambiguity. So I mined it myself. Weeks inside the platform — IT service management, IT operations, portfolio management, HR service delivery, and a handful of custom apps nobody had really measured before. What came back wasn't a slide deck of guesses. It was a prioritized set of agentic AI use cases, each with its own quantified value case — not aspiration, arithmetic. The same stakeholders who were guarded in week one are now asking how we extend the engagement. That's the part of this work nobody talks about. The value is rarely sitting in a requirements doc. It's sitting inside the platform, unmeasured, until someone goes and finds it — then proves it in numbers a CFO will accept. It's also a fair snapshot of the market right now. A small number of enterprises are genuinely operationalizing AI at scale. Most are one honest platform audit away from knowing where their own value sits. Genuine question for anyone doing this kind of work: how much of the "AI value gap" is actually just a discovery gap? Check the first comment for a detailed engagement story link #AppliedAI #AgenticAI #ForwardDeployedEngineering