Organizations are at drastically different places in their AI journey, and it’s important to point that out before prescribing anything.

Change is always hard at big organizations. Everyone is at a different point: Some senior executives have had the revelation about AI’s potential, while others haven’t. Senior vice presidents may have nodded agreement at the sound bites, but they haven’t spent a quiet Saturday morning immersing themselves in the tools. Maybe one in four have had their “this is it” moment with AI, while the rest are playing compliance theater. Some are fully committed but deeply frustrated: They believe in smaller, leaner, faster, better, but they genuinely struggle with how to get there without the wheels coming off.

We see four stages in this journey of making AI-first the new normal, noted in detail below. No enterprise has truly transformed to the last stage. Getting one step further from wherever you are today has a lot of value on its own. Ultimately, success depends on one thing: adaptability. The pace of change won’t slow down. Throw your teams into the deep end with the right scaffolding, and learn.

But first, a set of non-negotiables: conditions without which everything stalls. Then we cover the four stages: what each looks like, how to unlock value. Finally, we outline the real questions leaders are wrestling with and what it all adds up to.

The non-negotiables

These are not nice-to-haves. If they are not in place, everything downstream stalls. Speed of 50 to 100, or a speed of zero. 

1. Your top team has to feel it, not just fund it. You want your SVP-and-above rank excited because they have found the most interesting problem of their career. If your leadership is looking at this as “I don’t have another big transformation in me,” you are in trouble. This one isn’t delegable. Create a visceral “wow!” moment for every senior leader.

2. Own it from the line, not just the center. Line leaders own the AI productivity targets and show their teams what working differently looks like. And they work towards the central agenda: vision-setting, shared metrics, and forcing functions that make the frontline owner show up to the monthly CEO and quarterly board meetings. The center sets the bar and removes blockers. The line delivers.

3. Measure what matters, loudly. Dashboards showing adoption by team, by leader. Start measuring something, even if imperfect. Have your leaders directly broadcast wins with specifics. Celebrate breakout successes. Build frontline conviction with the real influencers your teams trust because if that person is using AI and is vocal about it, a team of 50 follows. 

4. Rewire IT, support, and approval councils for weekly speed. Your governance processes, security reviews, and vendor approval councils were designed for when you had years to change. You do not have years. If your approval cycle is quarterly and the technology cycle is weekly, you fall behind every sprint. Speed is non-negotiable: white-glove onboarding, champions in functions, tight feedback loops.

If you are running a transformation and the people in the room every week are not fired up, you either have the wrong people or the wrong conditions.

5. Have the right people, and protect them. Transformation runs on a small number of people who are genuinely willing to change and pioneer new ways of working. Sometimes they are not the most senior but the next generation who live on the front lines, connecting target-setting to workflow redesign to technology. Identify them. Shield them from legacy norms, approval chains, and red tape. Give them room to run. If you cannot find those people, or run out of them, pause. The best transformations have a palpable energy. The teams are excited, frustrated, pushing through walls, arguing about the right way to do things.

6. Win the hearts and minds of your A players. This is not just a cost exercise. You are unlocking the capacity of your people to do the work that matters. Make them feel it. 

The four stages mobilizing the organization

Every step in the journey from “we have AI tools, and we love it” to “we operate fundamentally differently” unlocks real value.

Stage 1: The light switch

What it is: Across your organization, maybe about 5% to 10% of people have had the moment. “I’ve met my match.” “I just did something I couldn’t do last week. Again.” They’re using AI constantly, telling their teams, pulling in their mentors, sharing examples with anyone who will listen. Tools are available. Most people have tried something, but the vast majority have not had the moment where it clicks. They have not sat down with a real deliverable and experienced what is actually possible now.

Value unlocked: up to 5% blended productivity, almost all of it driven by the small group who had the revelation, pulling up the average for everyone else. That concentration tells you who your change agents are. More importantly, you get the map: who are the early adopters, which functions have energy, where is the latent demand? The instinct is to worry about the 50% who are disengaged. Resist it. Accelerate the 5% to 10% and the next 20% to 40% who are curious and get it.

How to move the organization:

Create the moment. In your next town hall and your next SVP/VP all-hands, show everyone what’s possible and give them time to experiment. Offer concrete challenges; a blank prompt screen is a big hurdle to AI adoption. For example, draft a competitive analysis from scratch. Build a customer segmentation. Write a product spec. Build a working demo. Rewrite a sales playbook for a new segment. Allow a weekend or a week. Consider taking your teams offline for two days and have your top 5% teach the other 95%—hands on keyboard. Be prepared that half will love it and the other half will be lost. That’s fine. Make it an expectation. Get people to use the tools on real work. Everything else follows.

Stage 2: Raise the floor

What it is: First, you prove it works with a structured pilot with real teams, real metrics, and a real Net Promoter Score^SM. Give it 4 to 12 weeks, and success kills the debate permanently. Then you scale: Push the broader organization to use AI within their workflows with the same team structures, just faster. The floor for what normal productivity looks like has been raised across the board.

Value unlocked: as much as 25% productivity gain across adopting teams. This is consistent across every engagement. For a 10,000-person org at $150,000 fully loaded, 20% equals about $300 million in capacity freed annually. Some converts to headcount savings, some to velocity, some to backlog clearance, and yes, some to fund the AI tools and tokens themselves (see part two in this series, “How Token Economics Will Change Opex”). This is real money. There are changes to the workflows but not the operating model.

Continue to move the organization. Start with 10 to 20 teams in a structured pilot—not an exploratory feasibility study but real work: brownfield and greenfield projects, actual backlog items. Publish the results widely. Once the value is proven, scale with the non-negotiables at full force. Most of the organization can get here with tools, enablement, measurement, and senior leadership. It is also where most enterprises plateau. Stage 2 requires people to change how they work day-to-day, but it does not touch structural things: team sizes, reporting lines, role definitions. Those stay the same, but move faster.

Define the metrics that matter for each function. For development: adoption and usage depth, pull request (PR) velocity, feature throughput, mean time to recovery (MTTR), and ultimately engineering expense to revenue ratio (E:R). For sales: adoption and quality of use, increase in customer-facing time, pipeline coverage, conversion rate, and ultimately revenue per rep. The leading indicators tell you who is moving. The lagging indicators tell you whether it is working. You need both.

Propagate what works. As pilot teams learn, codify their end-to-end workflows, context files, agent configurations, and skills into reusable guidance. Do not let every team reinvent from scratch. This is how you scale from 20 teams to 200.

Be serious about extracting the value. You now know what the technology can do. Do not let it become just another cost of doing business. Define the extraction targets explicitly: efficiency, velocity, tech debt clearance, growth.

Watch for structural barriers: fixed incentive plans, compensation structures, quota setting, outsourced functions, long-term BPO contracts, agency relationships. These create dependencies that slow the transition and may need to be renegotiated.

Stage 3: Raise the ceiling

What it is: This often runs in parallel with Stage 2, but it’s your poster child. One leader with a meaningful organization commits to fundamentally changing how the work gets done. Not AI within the existing process. AI is the starting point for redesigning the workflow itself. Pods of 1 to 3 people instead of 15-person (or 100-person) teams. The frontier team is figuring out how to make developers 10 to 100 times more productive, marketers 10 times more effective, and sellers two to three times more successful. The structure changes, not just the tools.

Value unlocked: 2x to 3x productivity in the pockets where it lands. Five people refactor a product rather than 500 as originally expected. A three-person sales pod covers double the accounts. These are happening now. If 10% to 15% of your org gets here, the math compounds quickly: Costs decrease by 30%, and velocity doubles or triples in the functions that matter most.

How to move the organization here:

Pick the leader and the org: Find one SVP or VP who has personally had the revelation, runs a team of at least several hundred, and is willing to make their people uncomfortable. Give them a six-month mandate and direct access to the CEO and COO for blockers. Let them pick the top talent.

Clear the path and protect them. Ask them what is blocking them and remove it: red tape, approvals, procurement cycles. If it’s blocked on Tuesday, unblock it by Thursday. Protect them from the organizational antibodies. Peers will question why this team gets special treatment. Some leaders will feel threatened. The frontier leader needs explicit, visible cover. Without it, the antibodies win. And giving preferential treatment will demonstrate to the organization that this is the future—get on board.

Give them real flexibility. Flexibility to restructure teams. To change comp. To redefine roles. To hire and promote differently. If they are constrained by the existing HR framework and org chart, they cannot build the new model. The frontier team needs the authority to operate as if they were a start-up inside the company. Untether AI tools and encourage high levels of token usage aligned with increased productivity.  

The goal is not to reach Stage 3 once. It is to build the organizational muscle to re-reach it every time the frontier moves. That muscle is the moat.

Codify AI-first workflows. Already relevant in Stage 2, but it’s harder here because the frontier keeps moving. As the team figures out what works, capture it as institutional memory, reusable workflows, and technical scaffolding the next wave can adopt. This becomes company IP.

Stage 2 raises the floor. Stage 3 raises the ceiling. Every leadership team we work with struggles with where to put their energy.

Most are landing on a pragmatic combination: Let the frontier team run unconstrained, build the poster child, and simultaneously raise the pressure on everyone else to move—not gentle encouragement but real metrics and real accountability. The frontier inspires. The pressure ensures nobody mistakes this for optional.

Stage 4: The new normal

What it is: The frontier operating model scales to the full organization: agentic-first across functions. Has any large enterprise gotten here? No. The tools only became good enough in the last few months to even imagine it.

Value unlocked: the full waterfall (see Figure 1). Under this scenario, 70% of today’s opex is producing dramatically more velocity and value than the 100% before it. The companies moving toward this are building the adaptation muscle: Test each new capability, codify what works, roll it out, reset, repeat.

Figure 1

What else are leaders wrestling with?

These are the questions that come up in leadership conversations once direction is set, not afterthoughts. No one has definitive answers, but patterns are emerging.

Do we go gradually or put tension in the system? Option A: Transform gradually over two years. This is where most companies will feel most comfortable, but it’s not the speed of agentic AI. Option B: Reduce team sizes, keep targets, accept interim risk. Put tension in the system and raise the bar. This is how frontier companies and AI-natives are thinking about it because they see the imperative for speed and agility. They do not believe they have two to three years to get there, and neither should any leader.

How do we create the capacity to change? If AI is priority 6 of 10, you should not be embarking on this (or you should reset your expectations): You will not cross from Stage 1 to Stage 3. Some say change comes from doing everything “AI first,” but you also have to create room for people to actually change the way they work.

How do we tell real adoption from compliance theater? Every enterprise with an adoption metric has people gaming it: 30 queries a week, garbage questions, no behavioral change. Volume dashboards are necessary but not sufficient. How others are thinking about it: Measure quality alongside volume. Did the work product change? Did cycle time compress? Did they stop using the old workflow? If the answer to all three is no, the adoption is performative.

Who do we keep? The logical endpoint includes fewer people for the same amount of work. It is possible that fewer than you’d think have the disposition to thrive in an AI-first world. You know the top 5%. You do not know the rest with precision. The constant learners are the ones who will be indispensable: orchestrators, intent-definers, problem solvers. People who bring creativity, judgment, and trusted relationships. Say that out loud and mean it. Set the right expectations. Some of your best people will leave for start-ups because they see opportunities you have not created yet. The question is whether the ones who stay are the right ones. How others are thinking about it: Use real metrics, both leading (adoption depth, quality of use) and outcome (cycle time, output per person), to find the people who are clearly on the boat. After 90 to 120 days, the data sorts itself. Invest in the adapters. Have honest conversations with the rest.

In the past, it was either-or. Reduce headcount, or reinvest in velocity. Right now, smaller teams help you go faster. You get out of your own way.

What new talent do we need? This is an equally big and hard question. You need to create the opex space and recruiting muscle for new skills, in a market where every company needs those same people. How others are thinking about it: Reskill top talent into break-glass new roles. Recruit targeted profiles who blend domain expertise, operational chops, and willingness to work with the technology. Be willing to pay up. Every leader should be spending a few hours a week thinking about how they will fill the critical roles, making calls, doing interviews. You can’t outsource finding this talent. Some roles may not exist in your org chart today. Hiring for the right skills, cultivating a culture that attracts them, and aggressive performance management are the levers. Being an AI-first organization becomes a recruiting advantage in itself.

This is a rapidly moving journey: 5% change at Stage 1, 15% to 25% at Stage 2, and 2x to 3x at Stage 3. How fast you get there matters. The destination: 70% of today’s opex producing more than the 100% before it. Not because it is leaner. Because it is a better company.