At this point, I no longer think the important AI question is whether teams can build faster. They absolutely can.

But the real question still hasn’t changed, even now: do you know what is worth building?

Agents can now compress work that used to take days or weeks into hours or days. But the bottleneck does not disappear; it moves into the places organizations were already weakest: unclear goals, slow decisions, messy handoffs, disconnected leadership, and teams optimized for task completion instead of customer outcomes.

So, heads up:

• Engineers: parts of your job are quickly being automated. Not engineering itself. But the ticket-taking, boilerplate implementation, “just build what was written down” version of the job is getting far less valuable. The next version of engineering will require stronger product judgment, customer understanding, systems thinking, quality ownership, and the ability to use agents to move from context to working software.
• Product Managers: writing requirements was never the hard part. The hard part was knowing what problem was worth solving, understanding the customer and workflow deeply enough to make the right tradeoffs, and creating enough clarity for cross-functional teams to move in unison. Now, the hard parts have gotten harder: you are not just creating clarity for one workstream. You are now context-switching across several complex projects, all while keeping teams focused.
• Executives: ideas were never the hard part either. The hard part has always been judgment: knowing what matters, making clear tradeoffs, resourcing the team properly, and trusting high-agency people to execute. The market has never been noisier, and your team cannot move faster if they are guessing what matters. If you hired right, now is the time to give them the strategic clarity, resources, and trust to run.

So yes, AI agents can make your team faster. But they do not magically make your company more valuable. Instead, they expose whether your organization even knows what to do with more leverage.

The buffer was hiding the work

A lot of companies are not getting real value from agentic AI because they are trying to bolt it onto a way of working built for a slower era. We were at risk of doing the same.

Faster execution sounds great until you realize how much of your operating model depended on buffer. Product had buffer to think up the next feature. QA had buffer to test thoroughly and catch what slipped. Go-to-market had buffer to package and sell. Leadership had buffer to make decisions in longer cycles. But that buffer is rapidly disappearing.

Organizations can no longer hide behind process theater, status updates, or the comforting idea that “the work is in progress.”

While agentic AI has not removed the human operating model, it has stripped away the slack that protected the system from stress.

The first change was cadence, not technology

What changed for us was not “everyone use AI now.”

It was the model around the work: earlier prototypes, shared artifacts instead of static handoffs, tighter loops between Product, Design, Engineering, AI/ML, Data, and Operations, and more authority pushed to the people closest to the work.

We also stopped treating AI usage like a side hobby. We standardized workflows, outputs, and handoffs and made it part of how the team worked.

The results showed up quickly. Within the first couple of weeks, Product had folded prototypes into the PRD process, nearly all active projects had moved into the new workflow, and the Product team produced 3-4x more fully formed, compliant feature documentation and artifacts in one week. Over six weeks, we estimated the Product team’s overall operating speed at roughly 2x, with moments closer to 3x.

But the real lesson was not “AI makes documentation faster.”

It was that once the cost of creating artifacts drops, the quality of thinking around those artifacts matters more.

The unit of work got bigger

In a slower system, you can optimize inside functions. Discovery is one thing. Requirements are one thing. Design is one thing. Development is one thing. QA is one thing. Launch is one thing.

Once execution compresses, that stops being the right mental model. The relevant unit is now idea-to-launch.

So how do you go from idea-to-launch faster? I’ve found a lot of success by putting in the hard work to create a product discovery process that can move high-level strategic ideas out of the clouds and into tangible priorities, target outcomes, and phased rollout plans.

Yes, agents can help here too. They can explore codebases, generate prototypes, pressure-test workflows, synthesize context, and make ideas tangible much earlier for quicker refinement and smoother handoffs.

But agents do not remove the hard thinking. Not from Product Managers. Not from Engineers. Not from Executives. If anything, agents make strong judgment even more valuable, because making good decisions has become the slowest part of the system.

Shared artifacts beat perfect documents

Our biggest time savings did not come from one magical AI feature or agent skill. It came from reducing translation time between teams.

In the old model, one team described the work, another interpreted it, another built from that interpretation, and everyone discovered too late where meaning and customer needs had drifted.

In the new model, teams react to something tangible much earlier. That means less time rewriting requirements, less time debating what someone meant, less time cleaning up preventable misunderstandings, and a lot less development rework.

The same pattern showed up outside our Product team. Cross-functional teams working on operational bottlenecks moved faster because discovery, model changes, UI changes, operational feedback, and release decisions were happening in a much tighter sequence.

AI mattered because it shortened the gap between idea, artifact, and decision. But the bigger unlock was that the entire loop got tighter and clearer.

Progress reveals the next constraint

A faster operating model does not remove constraints. It reveals the next one sooner.

Once Product, Design, and Engineering sped up, QA got louder. Release management got louder. Instrumentation got louder. Cross-organization coordination got louder.

Annoying? Yes. Failure? No.

So we adapted. Release cadences became more dynamic. Grooming got more hands-on. Technical teams created their own skills to help streamline testing and deployments. We accelerated the business priority to automate our QA processes.

We didn’t shy from doing the hard work of changing the whole system. And it was only then that our outcomes started to change.

That is why I care a lot less about vanity metrics around AI usage and a lot more about whether the right work is moving faster. Token counts do not matter to me as much as the value we create with each launch.

What’s important to me is that cycle time is down, rework is down, decisions are made faster, releases are more frequent, teams are spending more of their time on work that changes the business, and customers are receiving more value than ever before.

The leadership job changed

Amid all these technical changes, the lesson I keep coming back to is leadership.

If development tasks get faster, leaders cannot keep managing the organization as a sequence of tasks and approvals. That just creates a new traffic jam around decision-making.

Teams need to be managed with goals and metrics, more than ever before. At this speed, there are too many moving pieces to manage by count alone.

The job of leadership is to make goals clear, make constraints explicit, give the team the right context and resources, and hire people you can trust to figure out how to get there.

That requires a different kind of management muscle: more clarity up front, more trust in execution, more accountability for outcomes, and less comfort in tracking every task as a proxy for progress.

If every meaningful decision still routes back to one human overseer, AI just gets the team blocked sooner.

At this pace, the best decisions increasingly have to come from the people closest to the product, customer, and user. Leadership still sets direction. Leadership still owns quality, compliance, security, and privacy. But it cannot orchestrate every detail from a distance and expect the organization to move at agentic speed.

This matters even more in healthcare. Moving fast does not mean being reckless. It means being much more deliberate about business, customer, and patient risks. It means being sharper about where oversight belongs and where to let your team run. And knowing the difference just became a differentiated leadership skill.

The real takeaway

Agentic AI did not make our whole company faster on its own. Changing how we worked together with these tools did.

AI agents shortened the distance between idea and artifact. They reduced the cost of iteration. They exposed weak handoffs. They raised the value of decision-making. They also made it much harder to hide a slow organization behind a lot of busy-looking processes.

Agentic AI did not replace hard work. It forced us to evolve the human operating model around the work.

Personally, I am not going to bet on the companies whose AI strategy is basically “use more AI.” I’m betting on the ones willing to rethink how they decide, build, validate, and ship around what AI agents make possible.

At the end of it all, what I keep coming back to is not whether teams are faster now. It is whether we are building something great for our customers.

Because in an age of AI agents, “can we build faster?” is no longer the most interesting question.

The question is: do you know what is worth building?

‍