When fintech giant Block announced 4,000 layoffs, founder Jack Dorsey pointed to AI-driven productivity as the reason. But reports from inside Block tell a different story: despite aggressive AI rollouts, engineers like Naoko Takeda saw little boost in productivity. CEOs may prototype with AI, but they rarely face the grind it takes to bring working code to production quality. The gap between bold claims and everyday results is yawning.

So what does this mean for you as a leader managing software teams? This article cuts through hype to show, using specific company examples and expert analysis, why human software engineers remain critical in any AI-driven transformation. We break down exactly where AI assists, and where it falls short, so you can make smarter decisions about your talent and tech investments.

AI Anxiety: Why Software Engineers Fear Replacement in 2026

The pace of change with AI in software engineering has fueled real anxiety among developers. Rapid gains in tools like Copilot and code generation models lead many to imagine their own roles shrinking or vanishing overnight. This fear persists even as new data shows that the reality of AI implementation is far messier and rarely lives up to sweeping claims.

Public statements by leaders, such as Jack Dorsey at Block, stoke these concerns further by implying that AI alone can drive drastic workforce cuts. On the ground, however, many engineers report that these narratives rarely reflect what happens in real software teams. Despite headlines, the day-to-day work still requires human logic, oversight, and the judgment AI cannot replicate.

The Reality Behind Headlines: Block’s Layoffs and ‘AI Washing’

Block’s pandemic hiring surge and financial pressures

Block’s headline-grabbing layoffs followed a pattern that had little to do with actual AI automation. The company massively expanded its headcount during the pandemic, then faced heavy financial strain as post-pandemic realities set in. Instead of acknowledging this, leadership leaned on AI as a convenient story. Announcing 4,000 cuts, Block’s founder Jack Dorsey cited AI “enabling a new way of working” and referenced new model capabilities as a driver.

The financial context tells the real story. Block’s pandemic-era hiring spree left it overextended when economic growth stalled and investor pressure mounted. AI became a scapegoat in communications, masking a basic cost-cutting imperative. This approach generates noise in headlines but tells operations and quality leaders little about the actual value or limitations of AI automation.

Employee perspectives on AI productivity and leadership decisions

The narrative from inside Block points to a gap between public claims and practical results. Data scientist Naoko Takeda reported that management “shoved AI down everyone’s throats” with minimal productivity upside. Her decision to reject a significant retention raise and walk away highlights frustration with leadership’s direction more than with the technology itself. Others at Block echoed skepticism, noting that bold claims overshadowed the real, incremental gains seen in daily work.

Aaron Levie captured this disconnect, pointing out that while leaders can build flashy prototypes, 90 percent of the effort in software engineering lies in pushing code to reliable production. The myth of rapid, total AI productivity ignores the reality of unfinished or unmaintainable output. For decision-makers, it is clear: using AI to mask business failings creates confusion, not ROI.

Decide-Execute-Deliver: Where AI Speeds Up, and Where It Falls Short

Hyperspeed in code execution vs. bottlenecks in decision and delivery

AI code generation tools like Copilot and large language models excel at the “execute” slice of the software development process, writing functions and suggesting syntax at speeds no human can match. Repetitive coding, boilerplate, and certain types of bug fixes happen faster and with less manual effort. This is the productivity spike, but it accounts for only part of the engineering workflow.

Two major bottlenecks remain: up front, in scoping and solution design, and at the end, in integrating, testing, and deploying in production. AI can automate the middle, but decisions on what to build, why it matters, and how to fit code into existing systems still require skilled human input. As noted in industry analysis, “AI compresses the ‘execute’ layer… but the other two layers resist automation in a way that will not be overcome by capability improvements alone.”

Why judgment and delivery layers resist automation

Engineering isn’t just about output. Decision and delivery stages deal with ambiguity, shifting requirements, and real-world constraints. AI models have no sense of business priorities, user needs, or company-specific architecture. Delivery means more than shipping code, it covers documentation, communication, compliance, and cross-team alignment. These layers demand context and judgment, qualities that no AI can replicate based on current technology or roadmaps.

Automation eliminates repetitive work, but it cannot substitute for the reasoning humans bring to critical tradeoffs and risk management. As the industry’s experience has shown, faster execution is valuable but creates its own bottlenecks if the rest of the process is not ready to keep up.

What CEOs and Teams Misunderstand About AI’s True Role

Overestimating quick prototypes vs. ignoring finishing work

Leaders see AI tools spin up prototypes in minutes and jump to the conclusion that full automation is at hand. There is a disconnect between the ease of assembling a flashy demo and the grind of production-grade development. As Aaron Levie noted in the source article, “CEOs are uniquely prone to delusions about AI’s usefulness because they can build quick prototypes but can’t see the 90% of work it takes to turn it into a finished product.” That missing 90% includes debugging, security checks, integrating with other systems, documenting, testing, and validating to meet specific requirements for uptime and compliance. Most of these finishing stages depend critically on human expertise and context, which cannot be automated by even the best code generation tools.

The gap between leadership vision and technical reality

There is persistent friction between leadership enthusiasm for AI and engineers’ day-to-day experience. Leaders want to believe that AI automation myths, like the idea that a new tool will almost instantly raise team productivity or slash headcount, are close to reality. On the ground, developers face the friction and unpredictability of real-world software environments. Staff see AI as a helpful assistant for routine code, not a replacement for complex reasoning, system design, or understanding requirements unique to operations and manufacturing. When teams are pressured to “AI everything” based on premature hype, they waste cycles chasing productivity gains that never materialize. The practical steps involve deploying AI on narrow, repeatable tasks, not handing over entire engineering workflows.

What This Means for Quality Managers and Manufacturing Leaders

Practical steps to assess AI’s fit for your team

Executives need to separate hype from day-to-day impact. Start by mapping your current engineering workflows against tasks AI tools can reliably automate. Routine, repetitive code writing is now faster with products like Copilot, but upstream tasks, problem scoping, validating requirements, still require expert judgment. Look for areas where your teams repeatedly lose time doing work that fits the “execute” layer described by Arvind Narayanan and Sayash Kapoor, rather than chasing use cases that demand full autonomy.

• Cut through “AI washing”: If vendors or stakeholders cite massive workforce reduction through AI alone, ask what real productivity data they have from your industry.
• Identify critical process handoffs: Find the points where quality depends on human review, validation, or adjustment. These gates are hard to automate, no matter the tool.
• Pilot, then measure: Commit to short, focused pilots. Quantify before-and-after results for actual code delivery, not prototype demos.

Setting real productivity expectations with AI tools in 2026

AI can compress parts of the coding process, but expecting “smaller and flatter teams” to deliver more without clear process redesign leads to disappointment. As seen in the Normal Technology analysis, large gains rarely materialize without significant investment in process realignment and change management.

• Productivity spikes are narrow: AI tools give best returns where tasks are well-defined and repeatable. Quality oversight still needs skilled people.
• Change requires patience: The 90 percent of work that turns prototypes into production is where bottlenecks remain. Plan for iterative rollout and realistic ROI timelines.

The goal: choose projects where AI augments, not replaces, your engineering talent. This approach defends job quality and protects ongoing productivity.

Looking Ahead: How Software Engineers Will Adapt, Not Disappear

Career adaptation and new job opportunities

As AI tools tackle repetitive coding tasks, software engineers will move into more specialized and strategic roles. Routine syntax and basic bug fixing are already less tedious, but someone needs to decide what features matter, set requirements, and guide architecture. This shift means engineers will increasingly focus on systems thinking, user needs, and overseeing complex integrations, tasks that AI cannot reliably automate. Teams will see more demand for roles like technical product managers and architects who bridge the gap between business goals and technical execution. New opportunities will also appear in areas such as AI model validation and oversight, ensuring that AI-generated code meets real-world standards before it hits production.

Demand outlook for experienced engineers

Forecasts continue to show healthy global demand for experienced engineers, even as entry-level coding jobs change shape. The move to compress “execute” tasks does not erase the need for people who can scope projects, design solutions, and check quality before delivery. As illustrated by the fintech firm Block, bold investments in automation do not reduce the complexity of shipping reliable products at scale. Companies will always need engineers who can work across teams, see hidden project dependencies, and maintain stability under pressure. Those comfortable learning new tools and guiding AI-powered workflows are least likely to face job risk. Adaptation, not elimination, is where work is headed.

Source: normaltech.ai