There was a time when the requirements document did not exist, because there was nobody on the far side of a wall to hand it to. The person who understood the problem and the person who built the solution were often the same person — one mind that could define a need, design the system to meet it, and trace how it worked end to end.
As systems grew more complex, we built professions and handoffs between those two people.
We treat the wall between business and technology as a permanent feature of any serious organization. It isn’t. It was a workaround — a rational response to a complexity problem — and the conditions that made it necessary are now reversing. AI is lowering the cost of interacting with systems, shrinking the distance between deciding what should happen and building it. The advantage shifts from specialists who live inside one domain to integrators who hold intent and implementation in a single short loop.
The mechanism matters here: easier interaction with a system changes how much translation is needed between a business decision and its implementation.
The Division Had a Reason
The separation wasn’t an error. As software grew and specialization deepened, understanding a system became its own discipline. In the 1970s, IT professionals sat embedded in business teams, with short feedback loops to the people who owned the problems [1]. The mainframe era of the 1980s and 1990s changed that. Dedicated IT units formed, handoffs were formalized, and the direct line from problem-owner to builder broke into a chain of intermediaries.
Each new specialization — data scientist, cloud architect, security analyst — deepened the divide, creating silos inside IT itself. Business defined requirements. IT implemented them. The requirements got documented, handed over, interpreted, and translated into systems that were then hard to trace back to the original intent. Coordination layers — analysts, project managers, governance boards — grew up to bridge the gap. They added friction without removing the disconnect underneath it.
The system worked. It was just expensive. And we made the classic mistake of confusing a temporary response to a technical constraint for a permanent organizational law — building whole career tracks to manage a gap that existed only because touching systems was hard.
So what happens when touching them stops being hard?
AI Lowers the Cost of Crossing the Gap
The defensible version of the claim is narrow, and worth the precision: AI lowers the cost of interacting with systems. Querying, configuring, and building become possible with far less technical mediation, and the threshold of expertise needed to close the loop between intent and implementation drops.
The numbers track it. Analysts projected that by 2024, roughly 80% of technology products and services would be built by people who are not technology professionals, using AI-assisted, low-code, and no-code tools [2] — with citizen developers expected to outnumber professional developers four to one in large enterprises. The research points the same way: Xu and colleagues find that expert-level AI deployment “uniformly lowers entry-level skill requirements, regardless of whether GenAI automates or augments” [11]. That reverses the specialization dynamic that justified separate, deep IT departments in the first place.
None of this erases technical expertise — I am not arguing the senior engineer is about to be obsolete. What changes is the dependency on translation: the layer that carries meaning across the business-IT boundary. The ability to shape a system moves closer to where the decisions about it actually get made.
If that mechanism is real, you would expect to see it in how organizations are built. You do.
You Can Already Read It in the Org Chart
Start at the top. 65% of CIOs now report directly to the CEO, up from 41% a decade ago [3]. A reporting line is an organization’s revealed belief about what matters — and this one says technology is strategic core, not service desk.
The role is changing shape underneath that line. 80% of technology leaders report significantly expanded mandates, and 66% of large enterprises treat their tech org as a revenue generator rather than a cost center. The boundary between “business strategy” and “technology strategy” dissolves at the executive layer first.
One qualification: these changes did not begin with the current AI wave. CIO reporting lines and product-based operating models have shifted for a decade, pushed by digital transformation and DevOps. AI is not lighting the fire here — it is pouring accelerant on one already burning, and lowering the floor for who gets to participate.
Lower down, the market is pricing the integrator directly. Listings for forward-deployed engineers — roles that fuse technical implementation with deep business context — rose more than 800% in 2025 [5]. And 57% of organizations are moving from project-based to product-based operating models, embedding engineers in business teams and cutting requirements-handoff stages [7].
Palantir built the cleanest version of this. Its AIP Bootcamp puts forward-deployed engineers and a client’s business team in the same room, on the client’s live data — no requirements document, no handoff. They iterate until a production workflow runs, collapsing a loop that used to take months into days. By mid-2024 it had run more than 1,300 bootcamps at roughly 75% conversion, and U.S. commercial revenue grew 71% year-over-year in early 2025.
Palantir works in defense, intelligence, and large-enterprise analytics, with budgets and talent density most companies will never match — the pattern transfers, the resourcing does not. But the pattern is the point. Remove the translation layer and you change not just how fast things get built, but what gets built at all, because drift accumulates at every handoff.
So Why Do the Coordination Layers Suddenly Hurt?
A layer that bridged a real gap becomes pure friction the moment the gap closes. By 2026, 20% of organizations are projected to flatten their structure and eliminate more than half of their middle-management positions [6] — precisely the roles that existed to manage the intent-to-implementation gap.
The compression is concrete. One healthcare company replaced a 10-person software team with a 3-person unit managing AI agents, and McKinsey reports 50 to 100 agents run by two or three people. When three people do what ten did, the handoffs between them aren’t optimized. They’re gone.
This is Conway’s Law working as an X-ray of the org chart. Organizations build systems that mirror their communication structures, so AI dropped into a divided org inherits the division — agents handing work to other agents along the same fractured lines where the humans already failed to talk. More than 80% of AI projects fail, roughly double the non-AI rate, with organizational silos a primary cause [8]. The robots, it turns out, are very good at reproducing our dysfunction faster.
The cautionary number sits underneath all of this: McKinsey finds only 6% of companies get measurable EBIT lift from AI, and what sets them apart is not how much AI they bought — it’s that they redesigned the underlying workflow [9]. The 6% rebuilt the loop. The rest, by and large, bolted AI onto the handoffs they already had. The variable is the org chart, not the model.
Moderna acted on exactly this. Two executive functions kept making decisions about the same workflows from incompatible vantage points, so it merged HR and IT into a single function — People and Digital Technology — under one officer, swapping separate “workforce planning” and “technology planning” for unified “work planning.” It now runs more than 3,000 custom AI tools, a reported 80%-plus of employees use AI daily, and roughly 50 digital-technology positions — about 10% of the combined function — were eliminated.
That last figure is the honest part of the story. Merging two executive functions and cutting 10% of a headcount is not a tidy reorganization; it is a disruptive bet. Moderna’s outcome shows what real redesign produces — not what renaming a reporting line delivers.
One condition determines whether the layers become unnecessary: people must actually be able to operate across the boundary. A shorter org chart alone does not give someone that capability. Otherwise the disconnect doesn’t vanish. It moves somewhere you can’t see it.
The Objections That Keep Me Honest
Three objections survive the optimism, and the strongest ones don’t have clean answers.
First: AI only lifts the surface. Even if the access layer converges, the decisions that matter most eventually touch distributed systems, security architecture, and compliance engineering — and that lower stack hasn’t simplified, it has grown more consequential. A leader who can now prototype a workflow with an AI assistant still cannot safely redesign a core banking ledger, and pretending otherwise is how you end up explaining an outage to regulators. AI also spins up its own translation roles — prompt engineers, model evaluators, AI auditors — fewer than the layers they replace, probably, but not zero.
So the translation layer compresses; it does not disappear. The move is to differentiate zones: collapse the layer where the loop has genuinely shortened, keep the specialists where complexity remains, but restructure them around integrators instead of intermediaries.
Second, and harder: maybe the integrator is just a rare, gifted individual, not something you can build an org around. The strong form of this isn’t a training-budget complaint — it’s a capacity ceiling. Holding genuine depth in both a business domain and an implementation stack may simply exceed what one head can carry, because AI handles syntax but not architectural judgment or the scar-tissue intuition that keeps complex systems from falling over at 2 a.m.
That holds only if integrating still demands 2015-level expertise in both domains. It doesn’t. An operator no longer needs to be a senior engineer to change a system on purpose when AI handles the implementation syntax and flags the constraint violations. But the honest tradeoff stands: that capability takes real investment — training, tool access, the political safety to work outside historic role lines. Skip it and you haven’t removed the bottleneck. You’ve just relocated it.
Third, the one with no clean resolution. Separation creates accountability and audit trails, not only translation. In finance, healthcare, and pharma, role separation is a compliance requirement, not a convention — collapse it and you transfer the compliance burden onto individuals without changing a single regulatory expectation. The thesis holds for layers that manage the complexity gap, not for layers that manage accountability. Telling those two apart is the first real skill of the convergence era, because “governance” is the word we use for both genuine compliance and “we would prefer not to change how we operate” — and on an org chart, those two look identical until somebody asks.
So How Do I Find the Gaps?
Convergence is something you build on purpose, not something the tools hand you. The Integration Gap Framework below is a diagnostic to run on your own operating model — a way to look, not a methodology to adopt.
- Map the translation layer. Find every point where business intent becomes technical spec. Count the roles between a problem and a working system, and mark which ones manage a handoff versus own an outcome.
- Assess integration capability. For each key role, ask: can this person define what should happen and grasp how it gets built? Pilot one direct interaction and watch where it breaks — it’s usually a knowledge gap, not unwillingness.
- Identify and protect your integrators. They already exist, working informally across the boundary, rarely recognized and almost never rewarded for it. That makes them a retention problem, not a hiring one — the kind of person who quietly leaves before you notice they were holding things together.
- Compress the feedback loop. Eliminate handoffs; don’t just speed them up. The test: can the person who framed the problem explain how it was solved?
- Re-anchor structure and incentives. Coordination layers live in budgets and performance reviews, not just org charts. Remove one, measure what breaks against what improves, and propagate only what survives.
One boundary applies to all five: in a regulated industry, without a sponsor accountable for both sides, or with no data on what your translation actually costs — assess before you restructure. Reforming a system you can’t measure isn’t transformation. It’s reorganization theater with a better deck.
The pattern under all five steps is the same. As systems get more accessible, the advantage shifts to people who can integrate perspectives rather than operate inside a single domain — convergence is less about roles than about how many perspectives one person can hold inside one shared system.
What to Do With This
Start where it’s cheapest to be honest. Tag every business-IT handoff you have as technical complexity, accountability requirement, or “we’ve always done it this way,” and attack that last category first — it’s the only pure liability, and you can find it by asking why each handoff still exists.
Then find one feedback loop where the round trip from business intent to working implementation takes longer than a week. That delay is almost certainly a translation-layer bottleneck rather than genuine complexity, which makes it your highest-return redesign target.
And if your AI initiative reports into IT with no mandate to touch the operating model, your structure has already answered the question this essay is asking — just not in your favor.
The divide between business and technology had a birthday. So it can have an end date. It won’t arrive on its own. But the conditions that created it are the ones now coming undone — and the people who notice first are the ones who get to decide what replaces it.
References
- [1] “The Evolution of IT Roles: From Business Partners to Silos, and How to Reunite.” LostConsultants, January 13, 2025. https://www.lostconsultants.com/2025/01/13/the-evolution-of-it-roles-from-business-partners-to-silos-and-how-to-reunite/
- **[2], ** “The Future of Citizen Development.” Quixy, reporting Gartner forecasts. https://quixy.com/blog/future-of-citizen-development/
- **[3], ** “Deloitte Tech Survey Reveals How Leaders Redefine Enterprise Value.” Deloitte Technology Executive Survey, 2025. https://www.deloitte.com/us/en/about/press-room/deloitte-tech-survey-reveals-how-leaders-redefine-enterprise-value.html
- [5] “The Palantirization of Everything.” Andreessen Horowitz (a16z), 2025. https://a16z.com/the-palantirization-of-everything/
- **[6], ** “AI Is Changing the Corporate Org Chart.” Fortune, August 7, 2025. https://fortune.com/2025/08/07/ai-corporate-org-chart-workplace-agents-flattening/
- [7] “AI and the Future of the IT Function.” Deloitte Tech Trends 2026. https://www.deloitte.com/us/en/insights/topics/technology-management/tech-trends/2026/ai-future-it-function.html
- [8] Shehu, Mohammed. “Conway’s Law for AI.” (Conway’s Law originally formulated 1968.) https://mohammedshehu.com/conways-law-ai/
- [9] “The State of AI 2025: Agents, Innovation.” McKinsey & Company, November 2025. https://www.mckinsey.com/~/media/mckinsey/business%20functions/quantumblack/our%20insights/the%20state%20of%20ai/november%202025/the-state-of-ai-2025-agents-innovation_cmyk-v1.pdf
- [11] Xu et al. “Expert-level AI deployment and entry-level skill requirements.” arXiv:2506.00532, 2025. https://arxiv.org/abs/2506.00532