Four New Resource Constraints on Your AI-Era Product Roadmap

The Scarcity Shift: AI made it easier to build almost anything. It didn't make it easier to afford, maintain, or justify everything.

For decades, nearly every roadmap meeting started with the same question: do we have enough engineers to build this? Uber’s recent, unexpected development challenge shows how little that question tells you about what belongs on your product roadmap in the AI era.

The company pushed its engineers to maximize productivity with AI coding tools, sidestepping the age-old capacity conundrum. There was even a leaderboard to encourage usage.

But just four months into the year, the team had burned through its entire 2026 token budget.

The toughest part? A budget is tight to raise when you can’t prove its impact. And according to Fortune, Uber’s COO said it was “very hard” to draw a line between rising AI costs and useful features for customers, adding that the link “is not there yet.”

This isn’t a cautionary tale about AI coding tools, but a sign that the thing product leaders have planned around for decades has suddenly stopped being the constraint. AI hasn’t eliminated scarcity in software development; it simply shifted it to other areas that determine what’s worth building, operating, and maintaining.

When Capacity Controlled the Roadmap

Capacity was used to prioritize the roadmap for us. Engineering effort was the scarcest resource of them all and served as the deciding factor for whether teams would “build or buy” new features. In turn, most bad ideas died of expense before anyone had to argue them down. 

AI coding tools changed that. A prototype that used to take a quarter can now be done in a weekend, leaving teams wondering: why not just build everything? But when the engineering effort required to spin up a new feature is at an all-time low, it no longer serves as the most reliable barometer for where teams should focus. 

Because as the Uber example shows us, AI didn’t eliminate scarcity in software. It moved it. 

Four New Constraints Reshaping Your AI-Era Product Roadmap

If the limit is no longer how much your team can build, product leaders need a new way to evaluate what belongs on their roadmap and what makes more sense to offload to a partner. Four constraints increasingly determine where teams should focus: attention, token budgets, ownership capacity, and customer value.

1. Attention

AI has empowered engineers to produce more. But it hasn’t enabled them to pay attention to more. 

Endpoint Clinical, whose software runs clinical trials, learned this the hard way. Its homegrown analytics and dashboards were static and hard to white-label. Customers routinely requested adjustments, forcing engineering to code custom permutations and delaying their own core product roadmap. 

The team called it a “report factory” loop: every new customer request became another ticket for the engineering team to juggle. Their limited attention was being funnelled into non-core activities, rather than the clinical trial technology customers actually pay Endpoint for. 

AI makes this trap easier to fall into because saying yes seems cheap. But your best engineers’ context, judgment, and focus haven’t scaled with the output. In fact, renowned scholar Herbert Simon warned in 1971 that "a wealth of information creates a poverty of attention.” Swap “information” for “code”, and you have what everyone in product is experiencing in 2026.

Every extra thread a team picks up pulls focus from the one that makes the product worth buying. In other words, when each engineer can do more, where you point their attention matters even more.

2. Token Budgets 

Budget has always been a constraint for product teams. But AI tokens have changed the unit of economics. A salary is fixed, and a software license is predictable. A token is metered by someone else, billed by usage, and grows with every engineer, every agent, and every customer who touches what you build.

It’s tempting to assume costs will go down as token prices continue to fall. According to Gartner, running a trillion-parameter model in 2030 will cost providers over 90% less than it did in 2025. But that doesn’t mean your bill will shrink. 

Agentic models need 5 to 30 times more tokens per task than a standard chatbot. When consumption rises faster than unit costs fall, overall inference costs go up. In August, Gartner named the effect the “Inference Paradox” and forecast that AI inference costs per agentic workflow will rise more than fivefold through 2028. As analyst Will Sommer put it, “Product leaders cannot rely on more efficient token economics to rationalize AI costs.” 

Additionally, inference costs are only part of the ongoing “run bill” required to maintain and continuously offer AI features. This sits next to the “build bill”, the engineering time and AI compute it took to create the feature. Both ultimately draw on the same budget, and are increasingly spent on infrastructure customers never see and won’t pay more for.

3. Ownership

Just like most teams fail to budget tokens beyond the “build bill,” every build decision carries an ownership commitment many teams don’t account for. The feature is what you see: the conversational interface, the agent, the demo. Most of what you’re really signing up for sits below the surface.

AI features raise the stakes because they’re never finished. For example, models constantly change underneath you, and every update means re-testing the answers your customers rely on. Gartner expects competitive AI products to require complex multimodal ecosystems that someone has to develop and maintain. 

The same team that’s supposed to build what’s next has to absorb that upkeep, so it deserves its own line item in your product roadmap planning, not a footnote under “maintenance.” That’s why we created the three-layer model for intelligent products. It treats each feature layer as a separate ownership decision, so you spend your limited upkeep capacity only where it sets you apart. 

  • Layer 1, your domain expertise and proprietary workflows, is the only place engineering effort compounds into an advantage, so it’s the one to own. 

  • Layer 2, the intelligent experience your customers touch, is best shaped and white-labelled,

  • and Layer 3, the governance, security and multi-tenancy foundation with no finish line, is the one to never own.

The question isn’t whether your team can build all of it. It’s whether you want to own everything, indefinitely.

4. Customer Value

Having the capacity to build something doesn’t mean your customers will adopt it. Every feature still has to earn its place. Will your customers understand its value, be satisfied when they use it, and feel comfortable trusting it? 

Faster and cheaper building doesn’t remove these filters. On the contrary, when building anything is possible, it makes them matter more. That’s the gap Uber’s COO described from the cost side: plenty of token spend and AI usage, but no clear line to better products and features that customers valued. 

EverHealth faced the same question. In a recent webinar, Chris McFaul, who leads product for its DrChrono EHR, said the debate over building its own reporting was never about whether the team could do it. He said anything can be built; “the question really was should we.” For his healthcare customers, trust was part of the work from the start: any new solution, he said, comes with “time to value and trust that you need to build.” 

His team isn’t treating its AI analyst as finished at launch, either. It plans to track adoption, engagement, and answer accuracy, and to keep asking, in his words, “what’s the value, ultimately, that it’s bringing to our customers?” 

Capacity might tell you what you can build, but customers tell you what is worth building.

Pressure-Test Your Product Roadmap

If scarcity has shifted, your planning has to evolve with it. This simple exercise can help guide you. Ask one question for each of the four constraints to see where your roadmap still overindexes on capacity:

1. Attention: What are your best engineers working on right now, and is it something only your company can build?

2. Budget: What does your most-used AI feature cost to build and to run, and what happens to that number if usage triples?

3. Ownership: For the last thing you shipped fast, who owns it twelve months from now, and what did they stop doing to make room?

4. Customer value: Can you draw a line from what you’re spending to something customers adopted and valued? If not, you’re where Uber’s COO was.

Write the answers down to identify the constraint(s) you’re not planning for and where you need to adjust.

Spend What’s Scarce on What Sets You Apart

AI changed what product teams can build. It didn’t change what they should build. When almost anything is buildable, the teams that win won’t be the ones who build the most. They’ll be the ones who focus their limited resources on what only they can build, and hand off the rest.

At ThoughtSpot, we love builders. Building with ThoughtSpot Embedded is how you offload Layers 2 and 3 of an intelligent product and buy back the focus for Layer 1, the layer that differentiates you.

Of the four new constraints covered in this post, tokens are the one most likely to get your boss’ attention first. Our guide, Token Anxiety: The Hidden Cost of Building AI-Powered Analytics, shows where AI token spend actually goes, ways to measure and minimize this spend more effectively, and how to point it at what sets you apart. Read it now →

Next in the series: If the limit is no longer how much your team can build, then where is the new bottleneck? Why your bottleneck has moved from building to deciding.