The Spreadsheet Moment: How Claude Code Will Reshape Adjacent Engineering Roles

The Spreadsheet Moment: How Claude Code Will Reshape Adjacent Engineering Roles

In 1979, VisiCalc shipped on the Apple II and quietly began eliminating 400,000 jobs that did not yet know they were in danger. The people who lost those jobs were not accountants. They were accounting clerks — the professionals whose primary value was the precision and speed of their arithmetic. The spreadsheet did not replace accounting. It replaced the execution layer of accounting, and in doing so, forced the entire profession to reorganize around the work that a spreadsheet could not do: interpretation, strategy, judgment, and the kind of advice that requires a human being who is accountable for the outcome.

That reorganization took roughly five years. The profession did not shrink — it grew. But the roles that survived looked fundamentally different from the roles that preceded them.

Software engineering in 2026 is approximately eighteen months into its own version of this transition. Claude Code is the spreadsheet. And the roles in most immediate danger are not the senior engineers who design systems — they are the adjacent roles that were always defined, implicitly, as "the people who do what engineers don't have time for."

What Claude Code Actually Does That's Different

Before assessing impact, it is worth being precise about the mechanism. Claude Code is not GitHub Copilot with better autocomplete. The distinction matters for understanding which workflows it disrupts.

Copilot and its first-generation successors operate at the line and function level: they predict the next token, suggest a method completion, or generate a boilerplate class. The developer remains the navigator — deciding what to build, structuring the approach, and assembling the suggestions into coherent software.

Claude Code operates at the task level. It reads your entire codebase, plans an approach across multiple files, executes changes, runs tests, interprets failures, and iterates — with the developer defining the goal and reviewing the result rather than guiding each individual step. Anthropic's 2026 Agentic Coding Trends Report frames the shift directly: "Software development is shifting from writing code to orchestrating agents that write code."

The concrete numbers make this tangible. Stripe deployed Claude Code across 1,370 engineers; one team completed a 10,000-line Scala-to-Java migration in four days — work estimated at ten engineer-weeks without the tool. TELUS reports 500,000 hours saved across its engineering organization, averaging 40 minutes reclaimed per AI interaction. Rakuten used Claude Code to implement a complex feature in a 12.5-million-line codebase over seven hours of autonomous work, with 99.9% numerical accuracy and no human code contribution during execution.

By early 2026, an estimated 4% of all global GitHub commits are authored by Claude Code, with projections suggesting that figure could reach 20% by year-end, according to the same Anthropic report.

The operative word in all of these examples is "autonomous." The agent does not suggest; it executes. That is a qualitative shift, not a quantitative one — and it changes which roles are most exposed.

The Spreadsheet Parallel

In 1983, Lotus 1-2-3 dominated business computing. By 1985, Microsoft Excel had arrived. By 1990, the U.S. had lost roughly 400,000 accounting clerk positions — roles whose primary function was manual ledger maintenance, arithmetic verification, and financial data transcription. These were not low-skill positions by the standards of their era. They required training, certification in some cases, and years of experience to perform reliably at scale.

What replaced them? Approximately 600,000 higher-order accounting roles — financial analysts, forensic accountants, strategic advisors, and the new category of "management accountant" whose job was to translate what the spreadsheet produced into decisions that a board could act on.

The profession did not shrink. It was reorganized around the work that the spreadsheet could not do.

The parallel to agentic coding is not metaphorical. It is structural. The roles most threatened by Claude Code are the ones whose primary value is defined by structured execution: writing boilerplate, running regression suites, transcribing business logic into documentation, maintaining CI/CD pipelines for codebases that are now being rewritten faster than those pipelines were designed to handle. These are the accounting clerks of software engineering — and the spreadsheet has arrived.

The critical question is not whether these roles will change. They will. The question is what replaces them, and how quickly individuals and organizations can make the transition.

The Adoption Data: Where Things Actually Stand

Before forecasting impact, it is worth grounding the discussion in what the data actually shows — including its contradictions. The Stack Overflow 2025 Developer Survey found that 84% of developers are using or planning to use AI tools, with 51% using them daily. AI tools are no longer an early-adopter phenomenon. They are the modal development environment.

But the same survey found that only 29% of developers trust AI tools — down 11 percentage points from 2024. The most common frustration, cited by 66% of respondents: AI solutions that are "almost right, but not quite." Debugging AI-generated code consumes significant time for 45% of respondents. METR's rigorous randomized controlled trial, published on arXiv in July 2025, produced a result that surprised the field: allowing experienced developers to use early-2025 AI tools increased task completion time by 19%. Developers predicted a 24% speedup before the study; they estimated a 20% speedup after completing it — despite the actual data showing the opposite.

METR's February 2026 update notes that more recent models likely produce genuine speedups, and that selection effects in the original study design limit its generalizability. The honest reading: early-2025 tooling did not reliably speed up experienced developers on complex tasks; 2026 tooling probably does, but rigorous RCT evidence at scale is still accumulating.

The Harness State of DevOps Modernization 2026, surveying 700 engineers and technical managers across five countries, found that 45% of very frequent AI coding tool users deploy code daily or faster. But 51% of those same frequent users report increased code quality problems, 53% report more security vulnerabilities since adopting AI tools, and 47% face increased manual QA and validation workload. Faster generation has outrun the infrastructure built to catch its errors.

Six Roles Under Pressure: A Ground-Level View

Junior Developers: The Training Pipeline Problem

The employment data for early-career software developers is unambiguous and sobering. Entry-level tech hiring at the 15 largest technology companies dropped 25% from 2023 to 2024, according to the Stack Overflow blog's December 2025 analysis. U.S. Bureau of Labor Statistics data shows programmer employment fell 27.5% between 2023 and 2025 — while software developer roles (higher-seniority, higher-judgment) declined only 0.3%. For workers aged 22–25 in high AI-exposure software engineering roles, employment declined 6% while rising 9% for workers aged 35–49.

The structural explanation is straightforward: the tasks that defined junior developer work — writing boilerplate, fixing isolated bugs, implementing specified features from a ticket, running manual test scripts — are precisely the tasks that Claude Code executes autonomously. The entry point to the profession has been removed, or at minimum, dramatically compressed.

IEEE Spectrum's analysis of entry-level job market data notes a resulting expectation shift: positions now described as "entry-level" increasingly require "two to five years of experience" rather than the historical one-to-two years. Employers, freed from relying on junior developers for execution work, are selectively hiring those who can operate at a higher starting level. 70% of hiring managers surveyed believe AI can perform intern-level work; 57% report trusting AI output more than intern or recent graduate output.

The deeper damage is less visible: junior roles have historically served as the profession's training pipeline. The gap year of writing plumbing code, failing in controlled ways, and being mentored through debugging is how senior engineers were made. Without it, the profession faces a compounding problem — not just fewer entry-level jobs today, but potentially fewer qualified senior engineers five years from now.

This is the most serious second-order consequence of the adjacent role disruption, and it is the one most organizations are not yet planning for.

QA and Test Engineers: From Execution to Strategy

Quality assurance is, in many ways, the role most directly in the crosshairs of agentic coding — and simultaneously the one best positioned to survive, if practitioners make the right moves.

The Harness 2026 report reveals the paradox directly: 47% of organizations with very frequent AI coding tool adoption report that manual QA and validation work has increased, not decreased. Faster code generation is producing more code faster than existing test infrastructure can cover. 51% of frequent AI users report increased code quality problems; 53% report more security vulnerabilities. The testing backlog is growing, not shrinking. At the same time, the work that defined traditional QA — writing regression test scripts for known flows, executing manual test plans against specifications, filing bug reports for straightforward defects — is increasingly handled by AI agents. The QA Trends 2026 report from Tricentis notes that 75% of organizations have identified AI-driven testing as a pivotal component of their strategy, though only 16% have successfully adopted it.

The test engineer who survives this transition is not the one writing test scripts. It is the one defining what should be tested, designing the coverage strategy, and interpreting the results that the automated systems cannot contextualize. QA is evolving from a function that verifies compliance with specifications into one that asks: "Does this software do what users actually need, in conditions the specification did not anticipate?"

Technical Writers: Decision Fatigue as the New Job Description

Technical writing is perhaps the clearest example of a role where AI has not eliminated work — it has transformed its character in ways that are quietly exhausting.

AI now generates the first draft by default. Every documentation platform, every IDE plugin, every agentic coding tool produces documentation as a byproduct of code generation. The technical writer who spent 60% of their time authoring content now spends that time reviewing, validating, correcting, and making editorial calls about content they did not write. The SoftServe Technical Communication 2026 report captures the cognitive shift precisely: "The real strain is not volume. It is decision fatigue." Writers bear responsibility for content they did not author.

The value proposition of technical writing is not shifting toward people who write well. It is shifting toward people who can assess whether something is right, make consequential calls under uncertainty, and design information architectures that serve users rather than merely describing software behavior.

DevOps and SRE: The Bottleneck Moves, Not Disappears

The Harness State of DevOps Modernization 2026 contains what might be the most important finding for understanding near-term DevOps role pressure: organizations generating code faster with AI tools are not experiencing proportional improvements in deployment quality or reliability. The "AI Velocity Paradox" is the report's central finding: 69% of organizations that use AI coding tools very frequently experience deployment problems regularly. Average incident recovery time for heavy AI users is 7.6 hours, compared to 6.3 hours for occasional users. The DevOps.com 2026 forecast puts it plainly: "AI agents are collapsing development cycles, automating testing, and reshaping CI/CD. DevOps leaders will face new pricing models, new metrics, and new accountability frameworks."

The bottleneck in software delivery has moved from writing code to everything that happens after writing code. The DevOps engineer whose role survives is the one who has moved from infrastructure scripting to infrastructure governance.

Product Managers: Closer to the Code Than Ever

Product management is experiencing a quiet capability expansion that carries its own risks. With Claude Code capable of implementing features from a description, product managers who understand enough to write clear specifications — or who can verify that a generated implementation matches the intent — are acquiring leverage that did not previously exist. The pressure on product management is not job elimination. It is scope inflation without proportional headcount.

UI/UX Engineers Who Code: The New Frontier Narrows

AI code generation tools now produce functional UI components from design descriptions or uploaded mockups with reasonable fidelity. What remains irreducibly human is the upstream work: deciding what the interaction should feel like, understanding why users behave counterintuitively in user research sessions, making aesthetic judgments that require cultural context, and advocating for accessibility concerns that are not captured in the specification.

What Stays Human: The Durable Value Layer

Across all six roles, a consistent pattern emerges. The work under pressure is characterized by: clear specification, reproducible steps, single correct output, and bounded context. The work that remains durable is characterized by: ambiguity, political complexity, multiple valid answers, and consequences that extend beyond the codebase. InfoQ's 2026 analysis of architecture in the AI era identifies the core constraint directly: "The choice between monolith and microservices may hinge on communication patterns, politics, or talent availability — none of which an AI can infer."

This translation function — from messy human reality to precise machine instruction — is the durable core of every professional role that will survive the current transition. It is what the forensic accountant did with spreadsheet output in 1990. It is what the senior engineer will do with Claude Code output in 2026.

The other durable layer is accountability. AI agents do not own outcomes. When the security vulnerability reaches production, when the documentation causes a user to misconfigure a critical system, when the deployment fails at 3 AM — a human being is accountable. Organizations that reduce headcount in adjacent roles without increasing the accountability infrastructure around AI-generated work are not becoming more efficient. They are accumulating liability that will materialize at the worst possible time.

How Teams and Individuals Should Adapt Now

The 6–18 month window is characterized by uncertainty, uneven adoption, and significant organizational inertia. But several adaptations are both tractable and high-value today. For organizations: For organizations: Rethink the junior pipeline before eliminating it. Invest in QA strategy, not QA headcount reduction. Build the AI accountability layer now — decide explicitly who is accountable for AI-generated code that reaches production, who reviews AI-generated documentation before it is published, and who owns the DevOps pipeline when an AI agent made the last change.

For individual practitioners: The spreadsheet analogy offers the clearest individual guidance. The accountants who thrived in 1990 were not the ones who refused to learn Excel. They were the ones who learned it immediately and used it to do things that hand calculation made impossible — and then built their value on the layer above. Learn to use Claude Code fluently before your organization mandates it. Use it to take on work that was previously out of scope. Then build your visible value on the judgment, strategy, and accountability work that Claude Code cannot do.

The Honest Reckoning

Transformative tools do not reorganize professions gently. The 400,000 accounting clerks who lost their roles to spreadsheets were not made whole by the 600,000 new accounting positions that emerged over the following decade. Individual careers do not transfer smoothly across a structural shift, even when the aggregate numbers look neutral or positive.

The same will be true here. Some junior developers who would have grown into senior engineers will not get the training pipeline that role requires, and the profession will have a pipeline problem in five years. Some QA engineers who built their entire expertise on manual test execution will find that expertise devalued faster than they can retrain. Some technical writers will experience the cognitive exhaustion of AI review work without receiving compensation or recognition commensurate with its difficulty.

These are real costs, and they deserve honest acknowledgment alongside the genuine productivity gains.

What the spreadsheet analogy ultimately offers is not comfort — it is orientation. We have been here before. The profession reorganized. The skills that survived were the ones that the tool could not replicate: judgment, accountability, stakeholder translation, and the ability to operate productively in conditions where there is no right answer, only better and worse ones.

The 6–18 month window is not the end of adjacent engineering roles. It is the period in which those roles are renegotiating their terms of employment with the tools that are reshaping them. The practitioners who approach that negotiation actively — learning the tools, identifying the judgment layer, building visible expertise in the durable value zone — will be in a structurally better position than those who wait for clarity that may not arrive until after the reorganization is complete.

The spreadsheet did not ask for permission. Neither will Claude Code.

References

1. Anthropic. 2026 Agentic Coding Trends Report. https://resources.anthropic.com/2026-agentic-coding-trends-report

2. Becker, J. et al. Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity. arXiv:2507.09089, July 2025. https://arxiv.org/abs/2507.09089

3. Harness. State of DevOps Modernization 2026. https://www.prnewswire.com/news-releases/harness-report-reveals-ai-coding-accelerates-development-devops-maturity-in-2026-isnt-keeping-pace-302710937.html

4. Stack Overflow. 2025 Developer Survey: AI Section. https://survey.stackoverflow.co/2025/ai

5. Stack Overflow Blog. AI vs Gen Z. https://stackoverflow.blog/2025/12/26/ai-vs-gen-z/

6. Stack Overflow Blog. Closing the Developer AI Trust Gap. https://stackoverflow.blog/2026/02/18/closing-the-developer-ai-trust-gap/

7. IEEE Spectrum. AI Shifts Expectations for Entry Level Jobs. https://spectrum.ieee.org/ai-effect-entry-level-jobs

8. Tricentis. QA Trends for 2026. https://www.tricentis.com/blog/qa-trends-ai-agentic-testing

9. SoftServe Technical Communication. Technical Writing Trends 2026. https://medium.com/softserve-technical-communication/technical-writing-trends-2026-lessons-from-a-year-of-ai-73e107390052

10. InfoQ. Where Architects Sit in the Era of AI. https://www.infoq.com/articles/architects-ai-era/

11. DevOps.com. Predict 2026: Why AI Will Force DevOps to Reinvent Itself. https://devops.com/predict-2026-why-ai-will-force-devops-to-reinvent-itself/

12. NPR. How the Electronic Spreadsheet Revolutionized Business. https://www.npr.org/2015/02/27/389585340/how-the-electronic-spreadsheet-revolutionized-business

13. METR. We Are Changing Our Developer Productivity Experiment Design. https://metr.org/blog/2026-02-24-uplift-update/

14. JetBrains. 2025 Developer Ecosystem Survey. https://www.jetbrains.com/lp/devecosystem-2025/