EXECUTIVE SUMMARY |
Every enterprise leader deploying AI tools is running into the same counterintuitive problem: the faster AI produces output, the busier their people get. Organizations are generating drafts, analyses, code, leads, and campaigns at machine speed—but humans still need to review, approve, integrate, and act on all of it. The bottleneck has shifted from production to judgment. |
Companies abandoning AI initiatives jumped from 17% to 42% in a single year, according to S&P Global Market Intelligence—and a primary reason is that organizations cannot absorb AI output fast enough to realize value. The "workslop" phenomenon—AI-generated content that looks polished but requires extensive human correction—is costing large organizations an estimated $9 million annually in rework, according to research from BetterUp Labs and Stanford. In software engineering, AI-assisted developers merge 98% more pull requests, but PR review time has increased 91%—the bottleneck didn't disappear, it moved downstream. The workforce doesn't shrink in this model—it transforms. Organizations need upskilled workers who orchestrate AI output, not more AI tools producing more output.
|
The organizations that will win aren't the ones deploying the most AI. They're the ones redesigning their operations around a simple truth: machines produce, humans decide—and decision-making doesn't scale the same way. |
|
|
|
|
The Human Bottleneck: Why AI Is Making Your Organization Busier, Not Leaner |
When AI outpaces the organization, efficiency becomes overwhelming |
I've gotten more done this year than in any year I can remember. Over 500 newsletters published across six publications in two and a half years. Research that used to take a full day now takes two hours. Drafts that consumed entire mornings appear in minutes. Analyses I never would have attempted are now routine. |
And yet I have a hard time finishing anything. |
The production isn't the problem. I can generate research briefs, first drafts, competitive analyses, and content outlines faster than ever. What I can't do faster is the part that actually matters: editing, evaluating, and fine-tuning. The work that turns raw AI output into something I'd stake my reputation on. That work—the judgment work—takes exactly as long as it always did. Sometimes longer, because AI-generated content has a way of looking polished while being subtly wrong, and catching those errors requires a different kind of attention than writing from scratch ever did. |
So my backlog grows. Not a backlog of things I haven't started—a backlog of things that are 80% done, waiting for the human pass that turns them from drafts into decisions. |
This isn't a personal productivity failure. It's a structural problem hitting every organization deploying AI at scale. And if you're running a team, a department, or an enterprise, the version of this story playing out in your organization is far more expensive than mine. |
The AI productivity narrative has been simple and compelling: automate routine tasks, free humans for higher-value work, and watch output multiply. Hundreds of billions of dollars in enterprise AI investment rest on this premise. But a growing body of research—and the daily experience of millions of knowledge workers—suggests the story is more complicated than anyone anticipated. |
The problem isn't that AI doesn't work. It works too well at producing output, and not nearly well enough at the parts that come next: review, judgment, integration, and action. The result is a new kind of organizational debt—a growing backlog of machine-generated activity that exceeds human capacity to process it. |
What Is the Human Bottleneck |
The human bottleneck occurs when AI accelerates production faster than organizations can accelerate decision-making. It manifests at three levels. |
At the individual level, workers using AI tools report feeling simultaneously more productive and more overwhelmed. A Wharton study identified this as the "AI efficiency trap"—a predictable four-stage cycle where time savings in one area immediately convert to increased expectations, filling schedules with exponentially more tasks. |
At the team level, the bottleneck shows up most clearly in software engineering. Faros AI's 2025 Productivity Paradox Report, analyzing telemetry from over 10,000 developers across 1,255 teams, found that developers using AI assistants complete 21% more tasks and merge 98% more pull requests. But PR review time increased 91%. GitHub's Octoverse report confirms the scale: 41% of new code is now AI-assisted, monthly code pushes crossed 82 million, and pull requests are 18% larger and more architecturally complex. |
At the organizational level, the pattern repeats across every function. Marketing teams generate 10x more content variations but lack the editorial judgment to determine which ones to publish. Sales teams receive AI-scored leads faster than account executives can work them. Finance teams get AI-drafted analyses that require extensive human verification before anyone trusts the numbers. |
How It Works in Business Contexts |
The mechanics of the human bottleneck follow a consistent pattern across industries and functions. |
Stage 1: Task acceleration. AI tools compress the time required for discrete tasks—writing, coding, researching, and analyzing. Individual workers report meaningful time savings, often 30-50% on specific activities. |
Stage 2: Volume expansion. Because tasks are faster, more tasks become feasible. Projects that would never have been staffed now get greenlit. The total volume of AI-generated work product entering the organization increases dramatically. |
Stage 3: Review and decision backlog. Every piece of AI-generated output requires human evaluation before it creates business value. This evaluation work doesn't compress the way production work does—it requires human judgment, context, and accountability. |
Stage 4: Operational debt accumulation. The gap between what AI produces and what humans can process grows wider over time. Organizations begin carrying a growing inventory of unreviewed, unapproved, and unacted-upon AI output. |
Research from BetterUp Labs and Stanford, published in Harvard Business Review, found that 40% of U.S. full-time workers have received "work-slop" in the past month. Each incident costs nearly two hours of rework. For a company with 10,000 employees, the researchers estimate this adds up to $9 million annually. |
A finding from Stanford captures the perception gap: knowledge workers at one Fortune 500 company reported feeling 20% more productive with AI tools, while objective measurement showed they were actually 19% slower. |
S&P Global Market Intelligence found that the share of companies abandoning most AI initiatives jumped to 42% in 2025, up from 17% the prior year. |
How to Implement Throughput-Aware AI Deployment |
The solution isn't less AI. It's redesigning operations to match the throughput characteristics of an AI-augmented organization. |
Phase 1: Audit your decision throughput |
Before deploying more AI tools, map where human decision-making is already constrained. For every workflow you plan to accelerate with AI, ask: who reviews the output? How long does the review take? What's the current backlog? |
Practical steps: |
Measure the ratio of AI-generated output to human review capacity in each department Identify the three workflows where the gap between production speed and decision speed is widest Calculate the "operational debt" accumulating in each
|
Phase 2: Build decision infrastructure |
Just as you built data infrastructure before deploying analytics, you need decision infrastructure before scaling AI output. |
Practical steps: |
Establish tiered review frameworks: not everything requires senior review Create AI-output quality standards so reviewers know what "good enough" looks like Invest in tools that help humans evaluate AI output faster, rather than tools that produce more output
|
Phase 3: Redesign roles around orchestration |
The most important workforce shift isn't replacing humans with AI—it's transforming human roles from production to orchestration. |
Practical steps: |
Redefine job descriptions to emphasize review, curation, and decision-making skills Train teams on evaluating AI output critically rather than just generating it Create new roles explicitly focused on AI output management
|
Key Success Factors: |
Measure throughput (output that reaches business outcomes), not just output volume Invest in review capacity at the same rate you invest in production capacity Accept that some AI-generated output should never be reviewed—it should be discarded
|
Common Missteps |
Measuring productivity by output volume instead of business outcomes. The most dangerous metric in an AI-augmented organization is "how much did we produce?" Teams can generate 10x more reports while delivering zero additional business value. |
Cutting headcount while scaling AI output. If you reduce the number of humans available to review, decide, and act on AI-generated work while simultaneously increasing the volume of that work, you create an impossible math problem. |
Treating the review bottleneck as a temporary problem. It isn't temporary. The Productivity J-Curve research from Brynjolfsson, Rock, and Syverson at MIT shows that transformative technologies require massive complementary investments in workflow redesign before productivity gains materialize. |
Deploying AI broadly before deeply. S&P Global's data shows that organizations succeeding with AI are focusing on specific, high-value workflows and building complete end-to-end support. |
Business Value |
The organizations that solve the human bottleneck first gain a durable competitive advantage—not because they have better AI, but because they can convert AI output into business outcomes faster than competitors. |
ROI Considerations: |
The real ROI of AI isn't in production savings—it's in decision acceleration. Workforce investment in orchestration skills compounds over time. Unlike AI tools (which competitors can license), a team skilled at evaluating AI output is a proprietary advantage. The cost of operational debt is real and growing. Every piece of AI-generated work that sits unreviewed represents wasted compute cost and missed opportunity.
|
Competitive Implications: |
The companies that will lead their industries over the next three years aren't the ones deploying the most AI models. They're the ones building the organizational capacity to absorb AI output at scale. This is a workforce strategy problem, not a technology problem. |
|
IN PARTNERSHIP WITH ALL THINGS AI |
|
|
All Things AI 2026 — March 23–24 | Durham Convention Center, NC
I produce the All Things AI Conference with my business partner, Todd Lewis, founder of All Things Open. We are committed to upskilling and aim to provide the most valuable and accessible expert-led workshops in the industry. Here's what's on tap in Durham in March. Workshops sold out in 2025. Don't wait. Check out all the workshops here. | Conference Pass — $199 — Tuesday, March 24. Full conference access, 50+ sessions across 4 tracks, networking events, and session recordings. AI for DevOps Workshop + Conference — $299 — Monday–Tuesday, March 23–24. Full-day hands-on workshop with John Willis (Author of the DevOps Handbook and co-founder of the DevOps movement) plus full conference access. AI for Business Workshop + Conference — $299 — Monday–Tuesday, March 23–24. Full-day hands-on workshop with Mark Hinkle plus full conference access. AI for Agents Workshop + Conference — $299 — Monday–Tuesday, March 23–24. Full-day hands-on workshop with Don Shin plus full conference access.
| Prices increase after March 17. Compare that to $1,000–$3,000+ at other AI conferences. | 🎟️ Get Your Tickets Today! → |
|
|
|
I appreciate your support. |
|
No comments:
Post a Comment