Why Coordination, Not Code, Is the Real Constraint on Shipping Software Today
For nearly two decades, leading product and engineering teams at companies like Amazon, Uber and Zillow, the productivity bottleneck that I consistently addressed was the speed at which code could be produced. If you wanted to ship faster, you hired more engineers, cut scope, or accepted more risk. Every planning meeting I ever sat in eventually collapsed into the same question. How long will this take to build?
AI ended that almost overnight. An engineer with a modern coding assistant can now stand up a working prototype before lunch that would have taken a full sprint two years ago. Yet when I talk with CTOs and engineering leaders, almost none of them tell me there is less work than before. Prototypes appear in hours, but decisions and information sharing are still crawling out on the same old timeline.
The bottleneck did not disappear. It moved. Most organizations are still trying to make an outdated operating workflow fit in a world where it’s no longer feasible.
Also Read: AiThority Interview with Gou Rao, co-founder and CEO at NeuBird AI
The 60% nobody budgets for
Asana’s Anatomy of Work Index found that roughly 60% of knowledge workers’ time goes to what the researchers call work about work. That is, status updates, alignment meetings, chasing approvals, and repeating the same context to different internal audiences (and now AI agents).
I did not need a study to believe that number. I watched it happen for years. Some of the most talented people I have ever worked with routinely lost half their week to coordination work that never produced a single customer-facing improvement. Not because they were undisciplined, but because the organization ran on information that only moved when a human carried it.
Here is the uncomfortable math when it comes to how AI has impacted software development: AI accelerated the 40% of the week that was already the fun part of creating, while leaving the other 60% of overhead completely untouched. We optimized the obvious thing and ignored the most expensive and important element.
Faster engines, same drivetrain
It gets worse. Speeding up execution without touching coordination does more than leave the problem in place. It compounds it.
Every unit of output creates coordination downstream. More code shipped means more decisions to communicate, more stakeholders to update, more documentation to change, and more chances for the plan to quietly diverge from reality. When execution was slow, coordination could keep up on human effort alone. Daily standups and quarterly roadmaps were crude, but they were the glue that kept everything moving along.
Now imagine a team where AI agents complete what used to be three weeks of engineering work in two days. Features are implemented, tests are written, APIs are updated, and pull requests are merged before anyone logs in.
The challenge isn’t reviewing the code. It’s understanding what those changes mean.
Which product goals moved forward? Which decisions were made along the way? Did implementation drift from the original intent? Did one change create downstream risk for another team? What customers need to be updated?
AI is making execution exponentially faster. The bottleneck is becoming the organization’s ability to maintain a shared understanding of what changed, why it changed, and how those changes impact the outcomes the business is trying to achieve.
We keep prescribing more of the disease
The industry’s answer to coordination pain has been remarkably consistent for twenty years: add another system that humans must feed. A new dashboard. A tighter ticket hygiene policy. A weekly status doc with a better template. Another standing meeting to make sure everyone reads the status doc.
Every one of those fixes fails the same way when AI is added to the equation because each one asks people to do more manual work to describe what they are already doing. When everything is automated, these turn into pure overhead. Nobody wants to update these systems, so nobody does. An outdated dashboard is worse than no dashboard because people lose trust, distract the team with questions, or, worse yet, make decisions based on it.
For most of my career, this was simply part of the workflow. Develop or change the software, log the changes, and update stakeholders and customers. The synthesis work, reading the commits, the tickets, the threads, the meeting notes, and turning them into an honest picture of where things stand, was something only humans could do.
That is the part that just changed now. The same class of AI that writes code can now do the synthesis. But only if two conditions hold, and this is where most organizations will get it wrong.
What actually changes the math
First, the intelligence has to be ambient. AI has to listen to the systems where work already happens and surface what has changed, what is at risk, and who needs to know, without anyone having to file an update. The moment a solution depends on people remembering to update it, they have rebuilt the old failure with a newer logo. Coordination has to become a background process rather than a job function.
Second, it has to show receipts. If an AI-generated status cannot be traced back to the commit, the ticket, or the conversation it came from, leaders will not trust it, and they should not. Trust is the entire adoption gate. A confident summary with no evidence is just a faster way to be wrong.
Get those two right, and the math genuinely changes. The hours your best people spend narrating work become hours spent doing it, and the people you hired for judgment get to exercise judgment.
Where to start on Monday
1. Measure your coordination tax.
For one sprint, count the hours your team spends in status meetings, writing updates, and re-explaining context to one another or to an AI system. Leaders who do this are consistently shocked by the number.
2. Stop solving coordination with more coordination.
If the proposed fix requires humans to maintain it manually, it will eventually fail. Make automation the default.
3. Build a shared source of organizational context.
Goals, product decisions, ownership, dependencies, customer feedback, and product changes should live in a shared, searchable repository accessible to both humans and AI. Whether that’s a knowledge graph, structured documentation, or another continuously updated system, the important thing is that context isn’t trapped in Slack threads or someone’s memory.
4. Redeploy the recovered time with intention.
Don’t let the hours AI saves disappear into more meetings. Invest them in talking to customers, refining strategy, making prioritization decisions, and evaluating results. As AI takes over implementation, human judgment becomes more valuable, not less.
The teams that get the most out of coding assistants will not be the ones that generate code the fastest. Everyone will have that. The winners will be the teams whose understanding of what they are building, why it matters, and whether every change is moving them closer to their goals.
Code stopped being the constraint the day AI learned to write it. Coordination is next, but only for the organizations willing to address the new bottleneck.
Also Read: AI and The Future of Work: Artificial Intelligence Is Expanding Organizational Intelligence Beyond Human Limits
[To share your insights with us, please write to psen@itechseries.com]
Comments are closed.