96% of product groups now use AI constantly of their workflows, based on our current survey of 379 product professionals. Zero, not a single one, reported going with out it.
The decision is obvious: we’re performed imagining AI’s potential. For product groups, AI has moved from pilot to manufacturing. It is the brand new baseline, not the exception.
However adoption is just the start. The true query is not whether or not groups are utilizing AI. It is how they’re utilizing it to create lasting worth with out dropping sight of what makes merchandise profitable within the first place.
At this 12 months’s AI Product Summit, 15+ leaders shared hard-won classes on navigating a world the place tempo has accelerated whereas fundamentals—buyer ache, experimentation, belief—nonetheless stay the anchors. AI is not about if, however how and how briskly.
Here is what issues based on them…
(Plus what we introduced: Productboard Spark, our specialised product administration AI agent).
Classes for Product Leaders: Turning Hype Into Lasting Worth
For product leaders, the problem is not simply adopting AI—it is doing so with out dropping sight of what makes merchandise profitable within the first place. Velocity issues, however solely when paired with buyer worth, disciplined experimentation, and rock-solid governance.
1. Buyer Ache + Differentiated Worth = Profitable AI Product Technique
Paige Costello from Figma now ships AI options weekly, however pace is not the one metric. When requested what modified from pre-AI product improvement, she was clear: “We’re constructing for 3 years from now as we speak. That’s the solely strategy to construct.”
The method stays constant throughout corporations:
-
Yashodha Bhavani from Field filters each characteristic via buyer worth, safety, and confidence earlier than transport: “We sit for a second and assume: who’re you making an attempt to serve? How are you including worth? If the easiest way so as to add worth is to be a part of a hyperlink within the chain, then that is the way in which it is best to go.”
-
Emily Silberstein from Instacart prioritizes core consumer challenges earlier than layering AI on high. She gave a concrete instance: “73% of our customers have a dietary want. Earlier than AI, that was an enormous downside—onerous to unravel with 17 million objects. With AI, we discovered 1.4 billion knowledge factors about dietary wants in our catalog.”
-
Dharmin Parik from Uber AI echoed this: “Be very clear by way of what you need to obtain. The secret’s to give attention to outcomes.”
The underside line? Begin with the ache level. AI is the “how,” not the “why.”
2. Experimentation Has Change into the New Organizational Norm
Velocity with out construction = chaos. The query now: can we transfer quick AND be taught quick?
James Evans (Amplitude): “Clients have a lot higher tolerance for experimentation with AI than pre-AI. They need to lean in with us—partially to discover ways to construct AI themselves.”
Luke Behnke (Grammarly) on why interplay design issues: “Most agentic AI looks like command-line period computing. Grammarly spent 15 years bringing AI to your fingertips with out making you immediate. That interplay mannequin is what excites me.”
Paige Costello’s caveat: Groups should make clear experiment targets or threat losing vitality: “Design a roadmap with small, medium, and huge bets. Get into prospects’ fingers weekly and be taught at totally different scales.”
3. Belief and Governance Are Non-Negotiable
However experimentation with out guardrails creates threat. As groups transfer quicker, the stakes get greater. That is why belief and governance aren’t elective add-ons—they’re desk stakes.
Netta Haiby (Microsoft): “Reliable AI means high quality, governance, safety, security, privateness, observability, and management. Construct belief in from the beginning.”
She warns AI presents a special problem than earlier tech developments: “It may generate code, take actions. We want to ensure the AI stays inside intent—that it isn’t breaking what we supposed.”
John Kucera (Salesforce): “With out transparency frameworks, enterprises will not scale multi-agent methods. You want clear exec sponsors, clear KPIs, and people staying accountable.”
Classes for Product Managers: Constructing AI Merchandise
Constructing differentiated AI merchandise requires daring imaginative and prescient, relentless high quality focus, and technical depth to information engineers. Our audio system within the Constructing AI merchandise session spoke to simply that…
4. Set your moonshot product imaginative and prescient in an AGI-world, then work backward
Aashi Jain (Google DeepMind): “Assume Synthetic Normal Intelligence (AGI) is obtainable throughout the subsequent decade, possible a lot sooner. Ask your self: what issues nonetheless matter?”
Her framework: Map hypotheses throughout three key dimensions—know-how, customers, and enterprise—all constructed upon a vital basis of accountability and security.
- Know-how: If AGI is feasible, what different technical capabilities will / must exist?
- Customers: How will habits change on this planet of AGI?
- Enterprise: Methods to make your small business mannequin viable? Will there be new enterprise paradigms?
- Accountability & Security: What safeguards are non-negotiable?
Hint backward out of your moonshot imaginative and prescient. “Figuring out milestones and main indicators helps make clear which capabilities are inside attain versus far out, and divulges the place your largest dangers are.”
On uncertainty: “There’s an excellent likelihood many people will get it fallacious, and that is okay. The purpose is to attempt our greatest now, so we have laid a considerate basis for when AGI arrives.”
5. Context Engineering Is the New Aggressive Edge
Imaginative and prescient with out execution = hallucination. Vikash Rungta (Alloi.ai, former Meta) brings it again to execution.
His thesis: Immediate engineering will not differentiate you, managing context will.
Why? LLMs are stateless. “If there’s one factor you’re taking away: LLMs do not have reminiscence. Each query requires the whole context.”
Your moat = three reminiscence layers:
- Brief-term: This session’s interactions
- Mid-term: Compressed insights (“reserving household journey to Italy, needs Florence, versatile dates”)
- Lengthy-term: Deep preferences (vegetarian, budget-conscious personally, prefers United)
Rungta warns in opposition to “dump all the things”: “Simply because you may have 10 million tokens does not imply it is best to use them. If we will not course of tons of recordsdata, brokers cannot both.”
The self-discipline: Compress insights. Isolate context by process. Feed the best data on the proper time.
6. These Who Outline the Evals Outline What Qualifies as “Good”
Aman Khan from Arize AI warns that when groups optimize for technical metrics like hallucination or retrieval, enterprise metrics get misplaced.
Khan is direct: “The those that write the evals and take into consideration the metric of what is good are the identical folks which might be defining the standard of the AI product within the first place.”
The answer: PMs and subject material specialists—not simply engineers—ought to label knowledge and outline analysis standards as a result of they perceive buyer high quality.
Three questions Khan says each workforce ought to debate:
- What number of human labels do we have to really feel assured in our eval system?
- What occurs when the eval is nice however the human label disagrees anyway?
- What occurs when the eval is nice however the enterprise metric goes down? Who’s accountable?
“If evals go however enterprise metrics fail, you may have elementary misalignment,” Khan emphasizes. “You must be sure you’re really measuring the factor that issues to your organization.”
7. The Suggestions Loop Between Manufacturing Information, Evals, and Iteration Separates Winners from Losers
However how have you learnt in case your evals really work? You may’t simply “vibe examine” your strategy to manufacturing. That is the place the continual loop turns into vital.
Ian Cairns from FreePlay outlined the three-stage loop AI merchandise want—and crucially, why conventional software program improvement does not apply:
-
BUILD (Iterate on prompts rapidly): “You most likely weren’t having to repeatedly experiment to see in case your product labored—you simply designed it, constructed it, and examined it. That does not work anymore.”
-
TEST (Validate with consultant datasets): Not simply glad paths—abuse circumstances, edge circumstances, the bizarre stuff customers really do.
-
Actual Instance: One S&P 500 firm’s VP of Engineering advised Cairns their highest-leverage ritual is easy: “Each different Friday, we spend two hours locked in a room with area specialists. We simply learn via logs. That cadence has turn out to be key to how we enhance high quality.”
-
OBSERVE (Monitor manufacturing actuality): “Construct evals bottom-up from actual failure modes, not top-down from theoretical KPIs,” Cairns urges.
He emphasizes the continual nature of this loop: “Individuals who get to manufacturing with out good evals at all times say they need they’d began sooner. This is not ‘design, construct, take a look at, ship.’ It is a loop you run weekly.”
Classes for Product Ops: Constructing the Product Working Mannequin
Product ops groups are uniquely positioned to make AI work at scale. They see throughout silos, perceive workflows, and may orchestrate the methods that free PMs to do their finest work. Our audio system within the “The Rise of the 10x PM” periods shared how they’re constructing processes and motions that assist future-proof their product org and enterprise.
8. Deciding Methods to Determine Is One of many Most Vital Issues You Can Do
Chris Butler from GitHub argues that the majority groups lack deliberate decision-making processes, which kills pace in an unsure AI period.
He outlined 5 levels each choice goes via:
- Identification: When does a call affect others?
- Discourse: Generate choices and analysis standards
- Determination: One particular person decides (generally a consensus, however ideally not)
- Communication: Share with stakeholders
- Studying: Separate course of high quality from consequence high quality
AI can increase every stage—detecting violated assumptions, simulating lacking viewpoints, manufacturing dissent—however people keep accountable.
Butler’s key perception: “Consensus-driven cultures loop endlessly between discourse and choice. Being intentional about your technique—like veto-based selections—accelerates all the things.”
9. Product Ops Should Join Tooling Throughout Silos to Unlock PM Productiveness
The bottleneck has shifted: It is not engineering—it is product managers drowning in instruments and context.
Ross Webb from Product Crew Success argues that PMs at the moment are the constraint.
His answer? AI brokers that synthesize knowledge throughout your complete stack.
Webb demonstrated a working system constructed with N8N and LangChain:
Enter sources: Productboard (priorities + roadmap), Linear (dev standing), PostHog (consumer habits), sentiment evaluation instruments
Outputs generated mechanically:
- Government abstract with 2 vital insights
- Well being scores throughout key metrics
- Prioritized suggestions (purple/orange/inexperienced)
- Implementation roadmap
Outcome: “Inside 5 minutes it is best to have insights and know precisely what your subsequent steps are. You do not have to spend hours aggregating knowledge—the agent recognized vital funnel drop-off in remark posting and prioritized it instantly.”
Webb’s message to ops groups: Give attention to saving PMs time, construct a compelling ROI case for executives, and shift from managing particular person instruments to orchestrating clever methods.
He explains the basis trigger: “Product managers would quite go to a burnout workshop than really go to the supply of why they’re burnt out. The supply is busy work. Use AI to automate that so PMs can give attention to the strategic work that is really scary, as a result of that is the place they create worth.”
10. Government Purchase-in and Time to Experiment Unlock AI Success
Matt Johlie from Relativity emphasizes that ops groups should automate routine work to reclaim strategic time—which requires air cowl from management to experiment.
His method: “You should safe air cowl out of your boss. In the event you’re getting indicators from the highest that it’s essential remodel the way you do AI, you want time to discover instruments, perceive their strengths, perceive their weaknesses, and uncover alternatives. You want time to have these aha moments.”
His workforce’s focus areas (stack-ranked by affect):
- Intelligence flows (product suggestions → insights → motion)
- Speedy prototyping capabilities
- Democratized knowledge entry
- Information administration
- Communication transformation
- Course of automation
The tenet: AI as co-pilot, not autopilot. “Co-pilot is deliberately distinct from autopilot. For product administration, co-pilot considering is required. Autopilot considering is unhealthy. People should be ready to defend any AI output.”
Product Growth Fundamentals Have not Modified (However the Execution Has)
The decision? AI is remodeling how we construct merchandise, however the fundamentals have not modified. Buyer ache, experimentation, belief, and clear decision-making nonetheless separate nice merchandise from mediocre ones.
What has modified is the pace of execution—and the necessity for brand spanking new working fashions.
Velocity is desk stakes. The groups that win will likely be those that:
- Embrace uncertainty
- Construct tight suggestions loops
- Empower their folks with clever methods
In an age the place AI compounds each pace and threat, the meta-process issues as a lot as the method itself.
Need to see these rules in motion? Try Productboard Spark, our AI-powered product administration agent that embodies the tight suggestions loops, context-aware help, and workflow integration our audio system championed.
