By Kortney Rupp, Director, Paragon by Riviera Partners
While you can give a team the authority to ship, it still needs someone who can tell whether the work is ready.
In our first two articles in this series, we discussed why AI’s business impact depends on how an organization is structured and how to design an AI-native operating model where teams own more of the work and make more of the decisions.
While that model only works if the people in it can direct AI-enabled work and make calls that used to pass through several layers of management, most hiring processes weren’t built to find those people. Here’s what to look for now, and how to update the hiring and career frameworks you rely on.
Building the capacity to execute
Our 2026 Future of Tech Leadership report shows where organizations are directing their hiring attention, with 54% of companies prioritizing individual contributors, 48% AI-relevant technical roles below VP, and 35% technical leadership at the manager or senior manager level as high priorities, compared with just 26% for C-suite leadership.
We call this the builder layer: the people who turn AI ambition into systems that actually run. If leadership decides what AI should do for the business, the builder layer determines whether it can. They build the models, pipelines, infrastructure, and safeguards that take an idea from pilot to production and then keep it working at scale.
In practice, this includes engineers, applied AI specialists, data professionals, platform teams, product leaders, and security experts. These people are spread across functions and levels, so no single leader is usually accountable for whether the organization has enough of them. As a result, a company can have a strategy, an executive sponsor, and promising initiatives while still lacking the collective capability to deliver any of it.
In recent searches for clients, we’re seeing an increased demand for AI platform engineers, applied scientists, and product and design professionals working in AI-native ways. The hardest roles to fill involve building the orchestration and infrastructure required to run many agents at scale, which differs substantially from building a single agent for a single task. It requires understanding how systems interact, how to evaluate performance, and how to control failures.
Most companies are hiring for the builder layer the same way they hired before AI: they screen for titles, years of experience, and, more recently, AI keywords on a résumé. None of these predict whether someone can make AI work inside a real business. The capabilities that matter now show up in how a person works, not in what their résumé says. Each role is changing in ways that traditional hiring signals miss:
- Engineers: Companies still test whether candidates can write code. The work increasingly depends on whether they can judge AI-generated code, tests, and documentation for architectural fit, security, and whether it actually solves the problem.
- Product leaders: Discovery and prioritization are still the job, but they’re no longer enough. Strong product leaders now define the context AI systems need, the quality bar they must meet, and how the team will measure performance.
- Managers: Leading a large team is often treated as proof of leadership. Now managers have to stay close enough to the technical work to judge output and resolve tradeoffs. A history of managing headcount says nothing about that.
Across all three, the role is shifting from functional expert to AI-leveraged problem owner: someone who can take an ambiguous problem, orchestrate the work across people and AI, evaluate what comes back, and know where human judgment still has to make the call.
That means changing how candidates are evaluated, not just who is targeted. Interviews need to show how someone actually works: give them AI-generated output to critique, an ambiguous problem to scope, or a tradeoff to defend. Companies that keep filtering on credentials will hire people who look qualified and still can’t deliver.
Recognizing responsibility
In an AI transformation, one of the clearest signs of a high performer is that their current job gets smaller. They automate the work they used to do, hand the routine to systems, and free themselves for harder problems.
Most leveling frameworks read that as decline: smaller team, lighter workload, less to show on paper. That is the wrong conclusion. Someone who removes their own work has shown the judgment and leverage the company should be promoting. Level people on what they can be trusted with next: the size and ambiguity of the problems they can own, the decisions they make without escalating, and the work they eliminated, not just the work they did. Two people at the same level may then lead very different teams, or no team at all, and compensation has to hold up when that happens. None of this works unless leaders have a bigger problem ready to hand over. Without one, employees learn that being good at their job is the fastest way to lose it.
When output stops tracking headcount, the interview has to change with it. Most loops still ask candidates how big their team was, how many people reported to them, and how long they held the title. Those answers describe the role a person occupied, not what they did with it. In our own candidate conversations, we ask about change instead:
- What did you automate, and what happened to the time it freed up?
- Where did you shorten delivery, and how did you measure it?
- What structural change did you drive that outlasted your involvement?
- What did you decide not to hand to AI, and why?
Strong answers are specific. The candidate names the problem, what they personally changed, and what the business gained. Weak answers drift back to team size, title, or “we” without ever saying “I.” Team size is still useful context, but it is the least informative thing a candidate can tell you.
This works in both directions. Leaders who expect more output per person have to recognize and reward broader individual contribution, or their best people will take that leverage somewhere that does. Candidates should expect the same questions and prepare to answer them with evidence, not headcount.
Making hiring capable of recognizing talent
Ask hiring leaders what’s slowing them down and the first answers are about supply: 47% cite an insufficient pipeline, 43% say they’re hiring too slowly, and 42% report inconsistent candidate quality.
The next set of answers points inward: 40% cite compensation expectations, 28% can’t move candidates through their own process efficiently, and 24% can’t reliably assess technical proficiency or AI fluency.
Those groups are connected. If a quarter of companies can’t measure the capability they’re hiring for, “inconsistent candidate quality” may say more about the interview loop than about the candidates. A slow process and an outdated pay band turn a thin pipeline into an empty one. Before concluding the talent doesn’t exist, leaders should ask how much of the shortage they’re creating themselves. Three changes can make the hiring process better suited to that type of talent.
- Rewrite your scorecard around outcome ownership, AI-native execution, rapid adaptability, systems thinking, and judgment and influence. Define what evidence demonstrates each capability so your interviewers assess candidates against a common standard.
- Test with real work. One question we find particularly revealing is: “Walk me through an AI-native workflow you recently built, step by step.” Listen for how they use AI to expand what is possible, how they handle failures, and which decisions remain theirs.
- Fix your hiring speed before increasing sourcing. The most common process mistake we see is allowing “culture fit” to justify more rounds with people whose calendars are already full. Additional sourcing does little if qualified candidates leave while the organization schedules another interview round.
Get the order right
Over this article series, we’ve discussed how the org chart was a rationing system for scarce human hours, why AI changed what’s scarce, and how the companies pulling ahead are redrawing how work runs and who they hire to run it. The final point is that these only work together.
Structure without the right people gives you autonomous teams making bad decisions faster than ever. People without the right structure gives you high-agency problem owners with no authority, who leave within a year and tell everyone they know why. Neither will show up in a board deck until it’s expensive.
The most common mistake we see is not a bad strategy. It’s doing the right things in the wrong order. Removing a management layer is visible, decisive, and easy to announce, so it’s usually where companies start. But companies that start there will likely spend the next year explaining why delivery slowed. Instead, it should be where they finish.
Here’s the order that works:
- Diagnose before you design. Score yourself on technology structure, leadership engagement, and governance timing. Find the exact handoff where work stalls. It’s rarely where the org chart says it is.
- Map one real workflow. One, not a 40-page AI policy. Mark what AI does, what people review, and who owns the result.
- Build the platform before you multiply teams. Otherwise, ten teams will solve model access ten different ways, and you’ve reinvented the hierarchy with extra steps.
- Bring governance into design. It’s the cheapest change on this list, and with the highest return. Legal will be just as demanding at the end, only now you’ll be rebuilding.
- Give one team real outcome ownership. Test it by asking them to name the last three decisions they made without escalating. If they can’t, the authority is decorative.
- Fix incentives and leveling. Do it before the new structure collides with the old pay model. The pay model always wins that fight.
- Remove layers last. Only where AI has actually absorbed the coordination work. Remove the coordination first, then the coordinators.
- Start with one pod, not a reorg. Run the sequence with a single pod inside one function. Track what it automates, where delivery speeds up, and which dependencies remain. Then use that evidence to decide how roles, pay, and staffing should change. A reorg based on a hunch is just a very expensive way to find out the problem wasn’t the number of layers.
Final thoughts
The first phase of enterprise AI rewarded the companies that experimented early. The next one rewards the companies that execute consistently. To stand apart, give teams the authority to make the decisions their work requires. Then hire, promote, and pay people for the judgment they bring to those decisions and the results they deliver.
Continue the series
- Part 1: How Organizational Design Determines AI Execution — why the faster AI makes the work, the more obvious it gets that the work was never the bottleneck.
- Part 2: Building an AI-Native Operating Model — the five steps for deciding what AI does, who holds authority, and where governance enters.
Frequently asked questions
What is the builder layer?
The builder layer is the people who turn AI ambition into systems that actually run. They build the models, pipelines, infrastructure, and safeguards that take an idea from pilot to production and keep it working at scale. In practice it includes engineers, applied AI specialists, data professionals, platform teams, product leaders, and security experts. Because they are spread across functions and levels, no single leader is usually accountable for whether the organization has enough of them.
Which AI roles are hardest to fill right now?
The hardest roles to fill involve building the orchestration and infrastructure required to run many agents at scale, which differs substantially from building a single agent for a single task. It requires understanding how systems interact, how to evaluate performance, and how to control failures. Demand is also rising for AI platform engineers, applied scientists, and product and design professionals working in AI-native ways.
How should companies interview for AI capability?
Change how candidates are evaluated rather than only who is targeted. Interviews need to show how someone actually works: give them AI-generated output to critique, an ambiguous problem to scope, or a tradeoff to defend. One revealing question is to ask a candidate to walk through an AI-native workflow they recently built, step by step, listening for how they use AI to expand what is possible, how they handle failures, and which decisions remain theirs.
How do you level people when output stops tracking headcount?
One of the clearest signs of a high performer in an AI transformation is that their current job gets smaller, because they automate the work they used to do. Most leveling frameworks read that as decline. Level people instead on what they can be trusted with next: the size and ambiguity of the problems they can own, the decisions they make without escalating, and the work they eliminated. Two people at the same level may then lead very different teams, or no team at all, and compensation has to hold up when that happens.
What questions reveal how a candidate actually works with AI?
Ask about change rather than scope. What did you automate, and what happened to the time it freed up? Where did you shorten delivery, and how did you measure it? What structural change did you drive that outlasted your involvement? What did you decide not to hand to AI, and why? Strong answers name the problem, what the person personally changed, and what the business gained.
What are the biggest AI hiring bottlenecks?
Hiring leaders first cite supply: 47% an insufficient pipeline, 43% hiring too slowly, and 42% inconsistent candidate quality. The next answers point inward: 40% cite compensation expectations, 28% cannot move candidates through their own process efficiently, and 24% cannot reliably assess technical proficiency or AI fluency. If a quarter of companies cannot measure the capability they are hiring for, inconsistent candidate quality may say more about the interview loop than about the candidates.
In what order should companies make these changes?
Diagnose before you design. Map one real workflow. Build the platform before you multiply teams. Bring governance into design. Give one team real outcome ownership. Fix incentives and leveling. Remove layers last, and only where AI has actually absorbed the coordination work. Start with one pod rather than a reorg, then use that evidence to decide how roles, pay, and staffing should change.
Kortney K. Rupp is a Director at Paragon by Riviera Partners. Connect on LinkedIn.
