About
I am a product executive focused on making enterprise AI work in the real world. My…
Articles by Ashish
-
Popular tools to analyze big data faster in Azure HDInsight
Popular tools to analyze big data faster in Azure HDInsight
HDInsight Interactive Query (also called Hive LLAP, or Low Latency Analytical Processing) is used to query data stored…
-
Zero ETL analytics with LLAP in Azure HDInsight
Zero ETL analytics with LLAP in Azure HDInsight
Sharing the recording from my Azure HDInsight talk at Data Works conference in Berlin…
Activity
8K followers
-
Ashish Thapliyal reposted thisWorking with ai tools is great, but It is even more fun to do this with others. Join us tomorrow to see what we have been cooking!Ashish Thapliyal reposted thisAI is changing how we work. But right now, too much of that work is happening behind closed tabs, one person and one agent at a time. Helpful? Absolutely. A little disconnected? Also yes. The thing is, AI gets a whole lot more useful when it has the same context your team does: the conversations, knowledge, and tools that keep work moving every day. What if AI wasn’t just something you used on your own? What if your whole team could jump in, contribute, and build together? Something new is coming. Join us live to see what happens when AI becomes part of the team. Here’s what we’ll cover: ▪️ Why Slack is the best home for multiplayer AI ▪️ How teams are already putting AI to work in Slack ▪️ Live demos you won't want to miss ▪️ A sneak peek at what’s coming to Slack We can't wait to show you what we've been building. 🗓 August 20, 2026 at 11 a.m. PT
-
Ashish Thapliyal reposted thisAshish Thapliyal reposted thisEducation has always been something I've believed can change lives. Recently, I learned about Lakshmi Chandra, who started teaching children under a metro bridge in Delhi after realizing many of them had no access to education. He saw that when children were given the opportunity to learn, they began to dream bigger and moved away from cycles of poverty, violence, and crime. His story really stayed with me. Through The Aware Project, I'm raising funds to support his school and help provide educational resources for these students. Every donation, no matter the amount, helps give a child the opportunity to learn and build a brighter future. If you're able to support this fundraiser or simply share it, I'd be incredibly grateful. 🔗 Donate here: https://lnkd.in/g8KaU6ux Thank you for helping make education more accessible. ❤️ #TheAwareProject #Education #GivingBack #CommunityImpact
-
Ashish Thapliyal shared this#Agentforce U.S. Department of Veterans Affairs has awarded Salesforce a $1.6 billion Agentic Enterprise License Agreement to transform how VA serves America's 17 million Veterans https://lnkd.in/gRcjWv9F
-
Ashish Thapliyal reposted thisAshish Thapliyal reposted thisSelf-improving agents are the next frontier: “The agents that win the next few years won’t just be ones with the cleverest foundation model,” I explain below, “they’ll be the ones that learn from their own outcomes.” Salesforce AI Research was among the industry’s first to demonstrate a reinforcement learning (RL)-like model for optimizing agents, called Retroformer. At Salesforce #AI, we’re not just testing new experiments with various Agentforce 360 design partners ‑‑ we’re dogfooding them ourselves! Learn more in our new 9-minute explainer about recursive self-improvement (#RSI) for enterprise agents: https://lnkd.in/gcWm5rJM
-
Ashish Thapliyal shared thisI'm hiring a highly technical Senior Product Manager, LLM Fleet Management to join my #Agentforce team in #Bangalore, India. We operate 100+ LLMs across 25+ regions, processing 33T+ tokens annually. I need a hands-on builder to own the capacity intelligence and economics of this massive fleet. This is not traditional PM role. You must: Write Code: Use AI Coding Tools like Claude Code/ Cursor to build your own automation scripts and data pipelines. Know AI Economics: Deeply understand TPM/RPM rate limits and PTU vs. PayGo tradeoffs. Scale Systems: Apply rigorous math to forecast and optimize trillion-token workloads. If you are an operator who wants to run the engine behind the #1 enterprise AI platform, come join us. https://lnkd.in/gG-kNk69 (Note: Please apply directly via the link. Due to the volume of interest, I won't be able to respond to individual DMs or inquiries).
-
Ashish Thapliyal shared thisI’m hiring a Director of Product Management to lead Agentforce Trust & Guardrails at Salesforce. This is a senior IC role focused on one of the hardest problems in enterprise AI: helping agents safely reason, use tools, access data, and take action in production, especially for regulated and security-sensitive customers. The role sits at the intersection of agentic AI safety, enterprise trust, runtime guardrails, prompt injection defense, tool-use risk, and platform-scale product strategy. I’m looking for someone deeply technical, customer obsessed, and comfortable turning real customer escalations into durable platform capabilities. Strong fit if you have hands on experience in AI trust & safety, agentic security, enterprise security, regulated SaaS, platform PM, or applied LLM safety. Please apply here: https://lnkd.in/gxJeCYH6 #ai #saftey #salesforce
-
Ashish Thapliyal shared thisTwo important takeaways 1. Instead of telling AI exactly how to do a task, organizations define the business outcome (e.g., increase sales or resolve cases faster). The AI continuously plans, acts, measures results, and adapts to achieve that outcome using enterprise data. 2. Enterprise AI isn’t just about smart models. The real value comes from continuously measuring performance, learning from outcomes, applying governance, and improving agents over time in a safe, auditable way.Ashish Thapliyal shared thisAgents are delivering real outcomes. They resolve the case, qualify the lead, draft the follow-up — real outcomes that move real work. With tools like observability, you can watch how well each one performs and make it better. And with loop engineering, agents do more than finish a task: they plan, check their own work, learn from the result, and adjust. That’s a big deal. Here’s the part I find interesting: those wins are agent-level. A resolved case, a qualified lead — each one matters. And each one is also one piece of a larger goal: whether the business is winning. Did customers actually get happier? Did the pipeline get healthier? That’s a different altitude. No single task, however well executed, can answer those questions on its own. The goal isn’t one agent nailing its corner — it’s sales, service, and the rest of the business moving together toward the same outcome. That’s the real frontier, not just loops that optimize individual agent outcomes, but loops that can optimize across the whole enterprise toward a larger business goal. And that only works if something around the agent can see the whole field and tell whether the business actually moved. That’s Salesforce. For twenty-seven years, our customers have encoded their goals into our platform. And all that CRM data, customer signals, automation, and analytics turn out to be the perfect infrastructure to translate tasks into goals that agents can optimize against over time. The agent-level outcomes are here. The interesting question is what happens when an agent can see past its own corner to the business goal those outcomes add up to. https://lnkd.in/gqD8uVbCAgents Run the Loop. Only Your Business Knows the ScoreAgents Run the Loop. Only Your Business Knows the Score
-
Ashish Thapliyal reposted thisAshish Thapliyal reposted thisAnyone can build an AI agent in an afternoon, but most won’t succeed on their own. The teams struggling to scale AI agents today are applying the traditional software playbook, where 90% of the work happens before launch. The successful teams are flipping that structure, to one where 90% of the work happens after go-live guided by crystal clear KPIs such as containment rate or meeting schedule rate. Scaling requires an “agent manager” inside the business with tools for continuously monitoring, tuning, and expanding agents over time the same way you give feedback to a new hire and coach for excellence. At Salesforce, we've already had 20,000+ real enterprise agent deployments and 3M+ conversations handled by our support agent alone. We’ve seen how quickly real-world edge cases, ambiguity, and complex intent reshape what “production ready” actually means. I recently sat down with Alex Xu at ByteByte Go to break down our pre-launch foundations and post-launch lessons that we learned at Salesforce after our agent deployments, including context engineering, choosing between probabilistic vs. deterministic execution, and building enterprise-grade guardrails. Check out the full breakdown here: https://lnkd.in/gVsjd4cmWhat Salesforce Learned from 20,000 Enterprise Agent DeploymentsWhat Salesforce Learned from 20,000 Enterprise Agent Deployments
-
Ashish Thapliyal liked thisAshish Thapliyal liked thisBuilding is a multi-player sport. Slack Code the one place for human, agents, ideas and code. https://lnkd.in/evZW23RC
-
Ashish Thapliyal liked thisAshish Thapliyal liked thisAfter 14 incredible years at Salesforce, I’ve decided to step away from day-to-day leadership as Chief engineering and customer success officer. I will continue to be an advisor to the CEO for the next year to fully transition my responsibilities. It is hard to overstate how much Salesforce has meant to me and my family. It has been the highlight of my career and the honor of a lifetime. I joined when we were a $2B company and the mobile revolution just started with the iPhone. We are going to end this financial year at more than $46B dollars as per our guidance. Since then, I’ve had the privilege of leading our technology, customer success, professional services, and South Asia business units over the years. Helping them grow and thrive has been a tremendous experience. A huge thank you to Marc and Parker for continually placing their trust in me over the years, and to the Board, Marc's Leadership Team, and my extended leadership team for their passion, partnership, and for living our values day in and day out. To my colleagues across every team: thank you for making this the best 14 years of my career. The send off at Laulima is something I will always cherish and I’ll always be cheering for this Ohana and congratulate and wish the new leadership the very best. To all our trailblazers , you are one of the main reasons Salesforce is successful and it has been a honor meeting you especially at TDX, DF, TruetothexCore, and a number of other events. And to our customers thank you for trusting us with your transformations. Your success has always been the driving force and watching you navigate change, especially the shift into the agentic era, has been the most rewarding part of this journey. Thank you to everyone who reached out personally with your messages of support and advice. I’m going to take some time to reflect, recharge, and plan my next chapter. With deep gratitude, Srini 🙏🏽
-
Ashish Thapliyal liked thisWorking with ai tools is great, but It is even more fun to do this with others. Join us tomorrow to see what we have been cooking!Ashish Thapliyal liked thisAI is changing how we work. But right now, too much of that work is happening behind closed tabs, one person and one agent at a time. Helpful? Absolutely. A little disconnected? Also yes. The thing is, AI gets a whole lot more useful when it has the same context your team does: the conversations, knowledge, and tools that keep work moving every day. What if AI wasn’t just something you used on your own? What if your whole team could jump in, contribute, and build together? Something new is coming. Join us live to see what happens when AI becomes part of the team. Here’s what we’ll cover: ▪️ Why Slack is the best home for multiplayer AI ▪️ How teams are already putting AI to work in Slack ▪️ Live demos you won't want to miss ▪️ A sneak peek at what’s coming to Slack We can't wait to show you what we've been building. 🗓 August 20, 2026 at 11 a.m. PT
-
Ashish Thapliyal liked thisAshish Thapliyal liked thisOne of the most meaningful moments before leaving for Germany was receiving congratulatory letters from my Senators and Congressman recognizing my selection for the 2026–2027 Congress-Bundestag Youth Exchange (CBYX) program. It was so special to spend time in Washington, D.C., meet the U.S. Department of State team, and officially begin this journey alongside the other CBYX participants. And now, somehow, I’m already in Germany! 🇺🇸🇩🇪 I’m incredibly grateful for the opportunity to represent my community and the United States as a Citizen Ambassador and to spend this year learning, connecting, and experiencing a new culture firsthand. A special thank you to Congressman Pat Fallon for the thoughtful letter and for recognizing this opportunity. I’m so excited for everything this year has in store! #CBYX #CitizenAmbassador #CulturalExchange #Germany #YouthExchange #PublicDiplomacy
-
Ashish Thapliyal liked thisAshish Thapliyal liked thisBold, slightly counterintuitive intuition about today’s transformers: hidden states are generically information-preserving -> exact prompts can be reconstructed from the representation (in principle): Consider the geometry. For two different input sequences to produce exactly the same hidden state, the model params have to satisfy some set of equalities. Those param configurations occupy a measure-zero subset of param space -- i.e. a randomly initialized transformer is *injective* with probability 1. (And, ordinary gradient-based training doesn’t generally move the model onto that degenerate set) Giorgos Nikolaou + Emanuele Rodolà formally prove this for the final token hidden state @ last layer (https://lnkd.in/gex4GeQn). But this causal structure suggests a broader interpretation if true: a hidden state is a fn of everything in its causal past, i.e. generically retains enough info to distinguish that entire history. So e.g. the representation @ token 5 in layer 7 isn’t merely a lossy “summary” of tokens 1-5. Under generic parameters -- it UNIQUELY identifies the upstream computation that produced it!Language Models are Injective and Hence InvertibleLanguage Models are Injective and Hence Invertible
-
Ashish Thapliyal liked thisAshish Thapliyal liked thisHate to announce that unfortunately.... I am here to blast your feed again with another W!! This weekend, we were incredibly grateful to receive the NVIDIA Champion's Choice Award at Spark Hack: Seattle! This was probably the hackathon with the widest variety of projects we’ve seen so far. People flew in to participate, everyone came from completely different specialties, and there was such a crazy range of ideas being built. There was SO much to learn and explore. We didn’t get the dub on the main track this time, but it was definitely a learning moment for us, especially when it came to the consumer side of what we were building and how important it is to create a pipeline that is actually intuitive and user-friendly. No matter how cool the tech is!!! Building off one of our previous hackathon ideas, we built STRUCT, a full end-to-end robotics pipeline. Basically: Type what you want to build in natural language → get a digital twin with a full CAD + PCB → train it to function in YOUR environment using RL + AR-generated human demonstrations → visualize and test it in your own environment through VR. And maybe someday through something like Meta glasses 👀 Super fun to build. (POV it took us 4 hours just to get the Meta Quest to open our project) It was especially cool getting to take ideas we had explored before and push them much further into something that connected robot design, simulation, training, and real-world interaction into one pipeline. Here’s the demo: https://lnkd.in/g4Duz4qG Huge thank you to NVIDIA, Acer for Business , Ascend, Jonathan N., Jen Haller, Kiana Steele, @Grace Gong, for putting together such a wonderful event, and to all of the organizers, judges, mentors, and other builders we got to meet throughout the weekend! Of course my goated team Andrew Zhao, Gagan Shiva Kumara, Suraj Shivakumar, & Advaith V.! And a lil shoutout to our goats who managed to take their spot in two categories: Khalid M.,Fuad Abdella,Agneya Tharun, Maurice Barksdale III LFGGG 🔥 + my honorary team for building cool shit!! (and letting me crash in their room) Andy Wang, Angela Yang, Shubh Malhotra, Daniel Chirakarn Till the next dub. See ya 🫡 — Sky #NVIDIA #BuildInPublic
-
Ashish Thapliyal liked thisAshish Thapliyal liked thisSalesforce is named a Leader in the IDC MarketScape:Worldwide Agentic Contact Center as a Service Platforms 2026 Vendor Assessment. Just months from launch of Agentforce Contact Center Get the excerpt: https://sforce.co/4zoLdUk #ContactCenter #CCaaS #IDC #Salesforce
Experience & Education
Patents
-
Rule evaluation for real-time data stream
Issued US12411830
Honors & Awards
-
Member of Microsoft Leader Bench (Hi-Potential Program)
Microsoft
-
Member of Microsoft Leader Bench (Hi-Potential Program)
Microsoft
-
Member of Microsoft Emerging Leader Bench (Hi-Potential Program)
Microsoft
Recommendations received
-
LinkedIn User
2 people have recommended Ashish
Join now to viewOther similar profiles
Explore more posts
Explore top content on LinkedIn
Find curated posts and insights for relevant topics all in one place.
View top content