Smarter AI Models Won't Fix Your Deployment | Maryam Ashoori, VP PM/Eng (IBM, Watsonx)
IBM's Maryam Ashoori breaks down why enterprise AI success depends less on smarter models and more on governance, architecture, and designing for agents at scale.
In this episode, we’re joined by Maryam Ashoori, VP of Product and Engineering at IBM’s Watsonx platform.
With a background that includes 2 master's degrees in AI, a PhD in Systems Design Engineering, and named on over 30 patents at IBM, she’s been on the bleeding edge for over a decade. Currently leading the charge on Agentic AI and AI Governance at IBM, Maryam is a bridge between the theoretical frontier of AI and the messy reality of enterprise deployment.
In this episode, Maryam shares:
Why AI has been stuck in pilot purgatory for longer than expected, and what you need to do today for a successful enterprise deployment
Shenanigans on the “biggest, best model” crowd, and why often a smaller, more focused tool is the right choice
How to build an agnostic architecture that can handle the realities of an AI world where models advance faster than anybody can keep up
1. From AI as experimental “toys” to enterprise products (3:30)
When ChatGPT went mainstream, companies rushed to fulfill their boards’ demands for an “AI plan” — whatever that meant. But few companies started with the right question.
“They basically picked an application, a solution looking for a problem that it solves versus it has to be the way around: how can I benefit from this technology?”
Why this matters: AI excitement doesn’t equal value. The shift from experimentation to production starts with defining the business problem, not retrofitting AI onto your roadmap.
2. Who’s accountable when an agent goes rogue? (7:10)
Where enterprises previously had a FOMO (fear of missing out) if they didn’t have an AI strategy, now they have a fear of messing up. They’re worried about being on the front page of the news for AI mishaps.
Maryam shares a question her client advisory board raised. It’s the one product leaders should be obsessing over:
“Something goes wrong in front of the users. Who is accountable here?”
Agents introduce autonomy, which introduces ambiguity. And ambiguity introduces risk.
Before shipping AI agents, leaders need clarity around accountability, guardrails, and governance workflows.
3. Latency will kill your AI product faster than cost (20:40)
It’s tempting to obsess over model cost. But Maryam reframes the conversation such that cost is a signal, and latency is the real user killer.
She shared an example of an insurance chatbot that added LLMs behind the scenes. The new response time?
40 seconds.
It was practically unusable.
Why this matters: No customer wants to wait 40 seconds for a chatbot response. Optimizing for compute isn’t just about margin; it’s about user experience, energy consumption, and architectural design.
4. The role of the product manager is changing (16:20)
Maryam describes how the traditional PM → PRD → engineer workflow is already blurring:
“The product manager with AI assisted coding, they code, they build. They have an army of agents that is at the prototype level now, but is gonna keep getting better every day.”
With that shift comes a new mandate:
“We need to do more thinking than doing because doing, we can pair it up with AI.”
Takeaway: The next generation of PMs won’t just write PRDs. They’ll design for agents, think in abstractions, and orchestrate systems.
Final thoughts
If you’re a product leader building with AI today, the question isn’t:
“Which model is best?”
It’s:
“Is our system designed to evolve?”
Links
Resources
Reinventing SaaS: Zuora’s AI Transformation | Karthik Chakkarapani and Shakir Karim (Zuora)
Linear’s Secret to Building Powerful AI Products | Nan Yu, Head of Product (Linear)
Chapters
00:00 Introduction
02:27 Maryam's product and AI journey
04:08 AI pilots, ROI, and the rise of agents
08:01 Production fears, security, and accountability
10:49 Agent reliability
16:35 AI PMs & observability: Measuring outcomes in non-deterministic agent workflows
20:12 Why compute optimization matters
24:49 IBM's model-agnostic architecture
29:14 Conclusion
What does LogRocket do?
LogRocket’s Galileo AI watches user sessions for you and surfaces the technical and usability issues holding back your web and mobile apps. Understand where your users are struggling by trying it for free at LogRocket.com.