AI has triggered both enthusiasm and anxiety in product circles. AI promises to transform workflows, slash costs, and create dazzling new experiences — but for product managers, the question isn’t whether AI is powerful. It’s how to manage such a fast-moving technology.
Few companies are grappling with this challenge more publicly than Upwork. As the world’s leading work marketplace, Upwork introduced Uma, its Mindful AI companion to streamline job postings, enhance freelancer proposals, and make the hiring process more efficient. Dave Bottoms, Upwork’s SVP and General Manager of Marketplace, has been helming this effort, carefully navigating the tricky terrain of AI integration.
But the Upwork story is only part of the picture. Sushma Kittali-Weidner, Chief Product Officer at Rheaply — a circular economy platform at an earlier stage of its growth — offers another view shaped by leaner resources and emerging markets. Interviews with each of them reveal six crucial lessons for any product leader looking to build AI into their portfolio.
1. Build for Change, Not Permanence
Traditional software product development prizes stability and scalability. But in AI-driven products, the ground beneath your feet is constantly shifting. The model that powers your functionality today might be obsolete six months from now.
Bottoms learned this firsthand. Upwork deliberately built its AI stack to be modular, with what he calls an “optionality layer.” This allows the company to dynamically select the best model for any task — whether it’s an off-the-shelf LLM like OpenAI’s, or a proprietary model fine-tuned on Upwork’s behavioral data. “What we think is the best model today,” Bottoms explained, “may not be the best model tomorrow.”
This approach mirrors what Sushma Kittali-Weidner has seen at Rheaply. In an environment of constrained resources, she’s had to avoid over-engineering AI features. Instead, her team builds for fast iteration, recognizing that both the technology and the market’s expectations are fluid. She has also found it useful to find ways to explicitly measure the user experience and overall impact of other key metrics before and after introducing AI features.
The lesson: Product managers need to architect AI products for adaptability, not permanence. Your competitive edge won’t come from locking in today’s solution but from how quickly you can swap it out when something better emerges.
2. Solve for Friction, Not Novelty
One of the most common mistakes product teams make when adopting AI is to chase what’s shiny instead of what’s useful.
At Upwork, Bottoms and his team didn’t start by asking, “What could AI do?” Instead, they looked at where users struggled. Clients were wasting time writing job posts. Freelancers struggled to draft compelling work proposals. Both sides were navigating unnecessary friction.
So Upwork deployed AI to reduce those pain points. The Uma companion now automatically generates job posts and work proposal drafts — not to show off, but to smoothly address the Jobs to be Done that define Upwork’s value proposition.
Kittali-Weidner sees the same pattern in other industries. Too many AI pilots are abandoned because they start with technology, not user need. “People are looking for magic, but not thinking enough about how AI can create efficiencies in existing processes.”
The winning AI products aren’t the flashiest — they’re the ones that quietly remove obstacles.
3. Keep a Human in the Loop
It’s tempting to dream of fully autonomous AI workflows. But both Bottoms and Kittali-Weidner caution against handing over the keys too soon.
At Upwork, AI might generate a job post or recommend a freelancer, but a human still makes the hiring decision. AI can translate documents, but someone still needs to edit for nuance. Bottoms estimates that even with Uma, “80% of the work can be automated, but the last 20% still requires human judgment.”
Kittali-Weidner sees the same dynamic at Rheaply, where operational realities require human oversight to ensure that AI-generated recommendations to digitize inventory and facilitate asset reuse make sense in the circular economy’s physical, logistical world.
For product managers, the lesson is to design AI as a collaborator, not a replacement. Automation is powerful, but human trust, creativity, and contextual judgment remain indispensable.
4. Learn on the Fly — and in Public
One of the trickiest things about AI is that you don’t get to ship it once and move on. AI-driven products evolve in public.
Upwork’s Uma companion is a case in point. The initial job post generator was an optional feature. But once data showed that AI-generated posts led to better outcomes, the team made it the default — all while continuing to refine and improve the system based on behavioral data.
Kittali-Weidner echoed the importance of iteration, stating that launching AI features as opt-in during initial releases would establish trust and encourage adoption by placing control in the users’ hands. In startup environments like Rheaply’s, experimentation is not a luxury; it’s essential. Yet she cautions that many companies suffer from “pilot paralysis” — a reluctance to move past testing toward real deployment.
The key is to treat AI like any other product capability: Ship it, measure it, improve it. Perfection isn’t the goal. Progress is.
5. Create User Trust
AI can introduce efficiency, but the balance between automation and trust is delicate. Bottoms describes how Upwork had to ensure that clients were comfortable with AI-generated job posts, initially giving them the option to edit and customize these posts. AI works best when it provides value without feeling like an opaque, black-box process that removes user control.
Kittali-Weidner highlights Rheaply’s ‘suggestions’ feature, which provides editable AI-generated drafts, effectively establishing user control and trust.
The lesson here is to ensure that your AI systems offer transparency and room for user customization. Trust is paramount, especially when dealing with sensitive or high-stakes processes like hiring. By allowing users to fine-tune and adjust AI outputs, product managers can maintain a level of trust while still benefiting from the time savings and efficiency gains that AI offers.
6. Think Beyond Product to Business Models
Perhaps one of the most exciting possibilities for AI at Upwork is the idea of multi-sided marketplaces. In the future, AI may not only assist freelancers in submitting proposals and clients in writing job posts but also act as agents that represent both sides, conducting interviews or even negotiating contracts. Bottoms envisions AI agents facilitating more complex interactions, significantly expanding what Upwork can offer its users.
The broader lesson for product managers is that AI has the potential to not just optimize current business models but create entirely new ones. As AI becomes more capable of handling nuanced tasks, product teams should think about how these advances could open new markets or reshape their company’s business model. The true power of AI lies in its ability to transcend mere optimization and serve as a driver of innovation.
The AI Product Manager’s New Mandate
The emergence of AI doesn’t rewrite the product management playbook — but it does demand a shift in mindset. To thrive, product managers must master AI fluency and “vibecoding,” enabling independent acceleration of early prototyping and experimentation. Product managers who succeed in this new era will be those who:
- Build for change, not permanence
- Prioritize real user friction over technological novelty
- Keep humans in the loop
- Iterate relentlessly based on real-world data
- Create user trust
- Think beyond product to business models
AI may be fluid and unpredictable, but one thing is certain: The product leaders who integrate it thoughtfully will shape the future of how we work.