The Org Chart Is the Product: Software As Case Study for What’s Happening in Intellectual Work

Jim

A Pattern Worth Recognizing

In 2005, newspaper financials looked stable. Circulation was declining gradually, digital advertising was growing, and the major groups were reporting steady revenue. Craigslist had been live for a decade. The structural disaggregation of classified advertising — which represented roughly 40% of newspaper revenue — was already complete. The financial statements hadn’t registered it yet.

What followed was not a technology story. The internet didn’t make journalism worse or news consumption less valuable. It made the aggregation model that justified the incumbent’s scale structurally unnecessary. The printing plant, the distribution network, the regional advertising salesforce — all of it optimized for a model the environment no longer required. The activity continued and expanded. The organizations built around the previous model contracted.

Software is in an analogous position now. The activity — building tools, automating workflows, creating products — is expanding rapidly. The organizations built around the previous model are showing the same early indicators the newspaper groups showed in 2005.

What the Indicators Show

Adobe’s CEO of 18 years announced his departure in March 2026 (archive) simultaneous with a genuine earnings beat. The stock had already declined 40% from its highs despite consistent financial execution. The market priced the forward thesis before the board acknowledged it.

This is not an isolated case. Across enterprise and creative software, the pattern is consistent: solid near-term financials, contracting stock valuations, leadership transitions, and restructuring cycles that consume organizational energy without closing the competitive gap. The restructuring signals awareness of the problem. It does not generate the velocity required to solve it.

The Adobe situation contains a single diagnostic data point more revealing than any financial metric. In 2022, Adobe attempted to acquire Figma — a browser-based collaborative design tool — for $20 billion. Adobe had 30,000 employees, $24 billion in annual revenue, deep relationships with the designer community, and complete category knowledge. A small team had built something Adobe could not replicate internally. That is not a resourcing failure. The resources were present in abundance.

The Detroit Parallel

Toyota’s advantage over US automakers in the 1970s and 1980s was not capital, brand, or distribution. Detroit had all of those. The gap was organizational.

Toyota’s Production System located problem-solving authority at the factory floor. A worker who identified a defect had both the authority and the obligation to stop the production line. Information about quality and process flowed immediately to the point where it could be acted on. The feedback loop between observation and decision was tight and real.

Detroit’s management structure filtered the same information upward. By the time a floor-level observation reached a decision-maker, it was late, aggregated, and shaped by the political interests of the layers it had passed through. The response was slow. The defects accumulated.

The quality and efficiency gap that followed took two decades to close partially. It never closed fully.

The mechanism in software is identical. The engineers and product teams with the most accurate read of competitive reality — what customers actually need, where the architecture is failing, what competitors are shipping — carry the least decision authority. Information travels upward through layers that distort it toward what senior leadership wants to hear. The distance between observation and decision is long. The competitive environment is moving on a cycle measured in months.

The Submarine Case

In 1999, Captain David Marquet took command of the USS Santa Fe, a nuclear submarine ranked last in the US Navy fleet on retention and performance. The submarine operates in an environment that makes information filtering structurally untenable — the consequences of slow or distorted decision-making are immediate and physical. Marquet’s approach pushed intent and authority downward, to the crew members with direct knowledge of systems and conditions. Instead of issuing orders to be executed, command defined objectives and trusted expertise to determine method.

Within two years, Santa Fe ranked first in the fleet. The subsequent decade produced a disproportionate number of officers who went on to command their own vessels.

The structure remained. Rank remained. What changed was where decision authority lived relative to where information lived. The two were aligned rather than separated by organizational distance.

This pattern — effective regardless of organizational scale — is visible in software where it exists. The companies that maintained it through growth tend to share a specific property: founders still running operations, with the organizational architecture built around the work rather than around the management of the work.

Performative Versus Architectural Adoption

Two adoption patterns are operating now in the same industries, sometimes in the same organizations.

The first injects AI into existing workflows. Velocity increases. The management structure that approved the adoption can measure the output volume. Code is generated faster, drafts appear faster, analyses are produced faster. Technical debt accumulates at the same speed. The organizational architecture is unchanged.

Amazon’s engineering leadership convened a mandatory review in March 2026 (archive) following a pattern of outages attributed to what internal briefing documents described as “Gen-AI assisted changes” with a “high blast radius.” One prior incident involved an AI coding agent that, given autonomous permissions, determined the best course of action was to delete and recreate a production environment — causing a 13-hour service interruption. Amazon’s stated response was to require senior engineer sign-off on AI-assisted changes going forward: a verification layer added reactively after the architecture failed. Separately, the company had set a top-down target requiring 80% of developers to use AI coding tools at least once weekly, tracked from above. The metric measures adoption, not output quality. These two facts — mandatory adoption from the top, verification bolted on after failure — are the generation-first pattern made visible.

The second path reverses the sequence. Human judgment defines requirements and verification criteria. AI generates candidates. Automated verification eliminates failures. Human review selects what remains. The focus shifts from production to selection — orchestration, context, and verification become the primary functions. This requires decision authority to live where the expertise is, not where the org chart puts it.

The first path is what large orgs can approve and manage. The second path requires the organizational architecture they don’t have.

What AI Surfaces

AI tools reduce the cost of generating output — code, analysis, documentation, design. Organizations where decision authority lives close to the work extract the productivity gain directly. The people with the expertise to specify what should be built, generate candidates, and verify what’s correct are the same people, operating in tight loops.

Organizations where generation is fast but specification and verification authority are diffuse produce output volume without the architectural coherence that makes output useful. The velocity is real. So is the technical debt accumulating at the same speed.

This is not a new problem AI created. It is an existing organizational condition AI has made visible by accelerating the consequences.

The Newspaper Analogy, Precisely Applied

What ended the newspaper industry’s dominant position was not a single competitor. No one company displaced the incumbent groups. Fragmentation across every revenue line simultaneously did — classifieds to Craigslist, display advertising to Google, sports coverage to independent sites, local news to social platforms. Each fragment was taken by a smaller, faster entity optimized for that specific function, without the overhead structure the incumbents required to operate.

The activity expanded. Total news content produced and consumed after the transition exceeded what the incumbents had produced at their peak. The aggregation model that justified their scale became unnecessary.

The software transition has the same geometry. No single AI company is displacing Adobe, Salesforce, or the enterprise software cohort. Fragmentation across every workflow segment simultaneously is. Each segment is addressable by a small team with access to frontier models, unencumbered by the organizational architecture the incumbents require to operate.

What We’re Watching

The layoffs across large software companies from 2024 onward are releasing senior engineers and product managers carrying domain expertise, customer pattern recognition, and technical knowledge of what problems exist and what has been tried. The tooling available to small teams has crossed a threshold where a handful of practitioners can now ship a competing workflow tool in weeks.

Amazon’s outage pattern is one early indicator of what this transition looks like from inside a large org simultaneously cutting headcount and mandating AI adoption. The most-recommended commenter (archive) in the Financial Times coverage of these incidents named the asymmetry precisely: productivity gains flow upward, failure cost lands on the practitioners who remain. That asymmetry is a structural feature of the adoption model, not an accident of implementation.

The financial statements of the incumbent software groups will remain stable for some period. This is what 2005 newspaper financials looked like. The structural disaggregation is operational before the income statement registers it.

The organizations that are performing — by the measures that matter, not only the measures that are reported — share the properties Marquet identified on Santa Fe and Toyota embedded in its production system: decision authority at the point of expertise, tight feedback loops between observation and action, and objectives clear enough that the people closest to the work can determine method without waiting for direction from above.

That organizational condition is independent of size. It is not independent of intent.

companion to When Code Gets Cheaper, More Gets Built