Think of a software organization as a river system. Ideas, decisions, and coordination flow from upstream, through a series of dams — constraints that control how much work actually reaches customers downstream.
AI has blown up one of those dams. Code that used to take days takes hours. Features that took sprints take afternoons. The coding dam is wide open.
So where's the flood of value reaching customers? Companies have invested heavily in AI coding tools, but most aren't seeing the downstream impact they expected. The river isn't flowing faster. Something else is holding it back.
That's because the coding dam was never the biggest constraint on the river. Upstream, there's a much larger dam — built from meetings, coordination overhead, slow decisions, and knowledge trapped in people's heads. That dam is still standing. And it's the one that actually determines how much value reaches your customers.
The industry is celebrating that it busted a dam. It busted the wrong one.
Where the Time Goes
The data tells a clear story: coding was never where the time went.
Developers spend only 16% of their time on actual application development¹. The rest? Meetings, planning, documentation, waiting for reviews, searching for information, and context switching between tools. At large companies, developers spend over 12 hours per week in meetings alone². The organizational dam isn't just standing. It's getting bigger.
DX's research puts a dollar figure on it: developers lose a full day every week to organizational inefficiencies. For a company with 1,000 developers, that's $18.5 million in lost productivity annually³ — not from slow code, but from slow everything else.
The coding dam was always one of the smaller dams on the river. AI blew it wide open, and the industry celebrated. But the water still isn't reaching customers any faster, because the massive dam upstream — the one controlling how teams coordinate, make decisions, and share knowledge — hasn't moved at all.
¹ Developers Spend Most of Their Time Not Coding: IDC Report
² For Developers, Too Many Meetings, Too Little Focus Time
³ DX State of Developer Experience Report
The Real Dam
So what does this upstream dam actually look like?
Meetings that don't need to happen. Decisions that take weeks when they should take days. Knowledge locked in people's heads instead of flowing through the system. Cross-team coordination overhead that grows geometrically with company size — at 50 engineers, you have 1,225 potential communication paths⁴, and that's before you add the product managers, designers, and stakeholders.
DX's research into what actually drives developer productivity found three dimensions that matter most: feedback loops, cognitive load, and flow state⁵. All organizational. None about coding speed. John Cutler puts it more bluntly — an uncomfortable percentage of all salary spend goes to context switching, handoffs, and meetings about coordinating dependencies⁶. And 78% of workers say meetings make it hard to get their actual work done⁷.
Atlassian's research captured the paradox: 68% of developers save over 10 hours a week using AI tools, but they lose an equivalent amount to organizational inefficiencies⁸. The time AI gives back, the system takes away.
⁴ Engineering Team Scaling Challenges
⁵ DevEx: What Actually Drives Productivity
⁶ TBM 243: Capital Inefficiency
⁷ Workplace Woes: Meetings
⁸ Atlassian State of Developer Experience 2025
Busting the Right Dam
So what does it look like to direct AI at the right dam?
Start with meetings — the single largest controllable drain on productive time. The fix isn't just "have fewer meetings." It's replacing low-value synchronous time with better asynchronous infrastructure. Status meetings become async video updates. AI-generated meeting summaries and action items flow into searchable documentation, so decisions don't get relitigated and knowledge doesn't require another meeting to reshare.
Then protect the time you've freed up. Developers switch tasks 13 times per hour⁹ and need 23 minutes to recover from each interruption¹⁰. Fewer meetings means nothing if real-time notifications fill the gap. Calendar tools that block focus time are a start, but they don't stop Slack from breaking flow during those blocks.
The frontier is agentic. A quality gate on meetings — AI that reviews calendar invites for clear goals and agendas before accepting, the way CI/CD gates code before it ships. A gatekeeper for attention — an agent that triages all incoming interruptions based on urgency and your current focus state, so better information flow doesn't just become more noise. These aren't science fiction. They're the logical next step.
But redirecting AI at the right dam isn't enough on its own. How you bust it matters just as much.
⁹ Rethinking Productivity in Software Engineering
¹⁰ The Cost of Interrupted Work: More Speed and Stress
Watch for Floods
We already have a cautionary tale. Look at what's happening downstream of the coding dam.
Teams using AI coding tools are completing more tasks and merging more pull requests — but PR review time has increased 91%¹¹. The downstream channels simply can't absorb the volume. Sixty-three percent of organizations say they ship code faster since adopting AI, yet 45% of deployments involving AI-generated code cause production problems¹². More water, more damage.
And it's not just the coding dam creating this flood. The industry is opening new tributaries — asking product managers and designers to code with AI tools, adding more water sources to a river system that was already overwhelmed. More volume, same channels.
Speed without system design doesn't create flow. It creates floods.
¹¹ The AI Productivity Paradox
¹² The AI Velocity Paradox
Build the Spillways First
That's the lesson to carry forward as we turn to the organizational dam.
The solutions have dependencies. Proactive knowledge agents that surface information to you throughout the day sound powerful — but without the flow gatekeeper to control when and how that information arrives, you've just added another source of interruptions. Another tributary into an already overwhelmed system. The order matters: build the channels and spillways first, then open the dam.
That's the principle. Don't just bust dams. Redesign the river system around them.
Redesign the River System
The companies that win in the AI era won't be the ones that code fastest. They'll be the ones that redesign the river system so speed is actually useful.
AI isn't failing us. It's exposing an organizational design problem that was always there — one we could ignore when everything was slow, but can't hide from now that one dam is wide open. In my last post I argued that AI should navigate before it accelerates. This is the other half: when you accelerate, make sure the river system can handle the speed.
Look at your own organization. Where's the real dam? And how can you redesign your system and thoughtfully bust it?
Sources Referenced
- IDC — Developers Spend Most of Their Time Not Coding
- Clockwise — For Developers, Too Many Meetings, Too Little Focus Time
- DX — State of Developer Experience Report
- Full Scale — Engineering Team Scaling Challenges
- Noda, Storey, Forsgren & Greiler — DevEx: What Actually Drives Productivity
- John Cutler — TBM 243: Capital Inefficiency
- Atlassian — Workplace Woes: Meetings
- Atlassian — State of Developer Experience 2025
- Rethinking Productivity in Software Engineering — Springer
- Gloria Mark — The Cost of Interrupted Work: More Speed and Stress
- Faros AI — The AI Productivity Paradox
- Harness — The AI Velocity Paradox