Surviving AI without drowning in more work

AI shifts productivity from execution to judgment

AI can remove hours of low-level work from a skilled employee’s day. The bigger change is what happens next. Employees spend less time producing first drafts, routine code, and basic analysis. They spend more time reviewing options, making decisions, and solving harder problems.

Keith Jones, a cybersecurity researcher with more than 30 years in the field, has experienced this shift directly. He uses Anthropic’s Claude to complete work that once consumed much of his day. “It really feels like I have a team behind the scenes, but what I have is Claude,” Jones said. He can now ask the system for three approaches to a problem and focus his attention on evaluating them.

This changes the nature of knowledge work. A developer using AI may generate several implementations in the time previously required to write one. An analyst can produce a first draft before manually assembling every input. The human role moves toward defining the problem, setting constraints, selecting an approach, and checking the result.

For executives, the key constraint is therefore shifting from production capacity to human judgment. Faster generation has limited value if employees cannot determine whether an output is accurate, useful, secure, and appropriate for the business. AI proficiency should include delegation and verification skills alongside familiarity with specific tools.

Companies should reflect that change in how they measure productivity. Output volume is an increasingly weak proxy for business value when AI makes content and code cheap to produce. Measures tied to cycle time, quality, customer outcomes, defects, and completed business objectives give leaders a clearer view of whether AI is improving performance.

The operating model also needs to evolve. Employees need clear rules for which tasks AI can handle, which require review, and which require direct human control. Training should focus on practical workflows and decision quality. Jones captures the useful mindset: “I look at AI like it’s another new tool, and I’m going to learn it like I’ve learned any other tool.” The executive objective is to turn that capability into consistently better work.

AI output requires a human quality-control layer

AI can generate code and text in seconds. Verification remains the bottleneck. Large language models can produce inaccurate facts, invented information, generic prose, and unnecessarily large code changes while expressing those outputs with confidence.

Low-quality AI output is often called “workslop.” In software development, it can appear as boilerplate code, excessive changes, or implementations that look plausible but contain defects. A hallucination is more serious: the model presents inaccurate or fabricated information as though it were valid. Both problems create downstream work because another person must identify and correct the errors.

Craig Shue, computer science professor and department head at Worcester Polytechnic Institute (WPI), gives a clear warning: “Don’t believe the machine is infallible.” He adds that an LLM can retrieve misinformation from the internet and incorporate it into an answer. A fluent response therefore provides no assurance that its underlying facts are correct.

Companies need verification built into AI workflows. Users can ask a model to provide sources and explain the reasoning behind an answer. For sensitive analysis, employees should give the model approved reports, transcripts, or data sheets and instruct it to work from those materials. This reduces the information space from which unsupported claims can emerge.

Software development requires stricter controls. AI-generated code should be reviewed and tested locally before it enters the main project. Teams should also constrain the scope of AI changes. A request to fix one function can otherwise produce changes across hundreds of unrelated lines, increasing review time and introducing additional risk.

The same principle applies to business content. Employees should own every AI-generated document, analysis, or recommendation they forward. Passing raw output to another employee simply transfers the validation cost downstream. At scale, that behavior can erase much of AI’s productivity benefit.

For executives, governance should match the consequence of an error. A low-risk internal summary can use a lightweight review process. Production code, financial analysis, security work, legal material, or customer-facing decisions require stronger validation. The critical management question is who remains accountable for the final output.

AI therefore changes where organizations spend their quality-control effort. Generation becomes faster. Review, verification, and accountability become more important. Companies that design those controls into the workflow can capture AI’s speed without allowing unreliable output to become an operational liability.

AI productivity can create more work

77% of employees in a multi-year Upwork study said AI increased their workload. Among workers who achieved the highest productivity gains, 88% also reported feeling burned out. These figures expose a management problem: faster content creation can increase the amount of work flowing through an organization.

AI sharply reduces the cost and time required to produce emails, reports, presentations, meeting transcripts, analysis, and code. That capability encourages employees to produce more. Every additional output creates work for someone else who must read, prioritize, verify, approve, or respond to it.

J.P. Gownder, vice president and principal analyst at Forrester Research, describes the problem clearly. “Is AI saving people time? The short answer is yes,” he said. “But people also are being overwhelmed with overproduced things. Everyone wants to look busy and they’re producing more, but not necessarily better. Managers have to push back on that or it’s not really saving you time.”

The impact extends into IT. IDC’s Future of Work 2026 survey found that 24% of IT workers identify increased workload as a top concern related to AI. Productivity gains can therefore coexist with greater pressure on employees. The volume of generated work matters because human attention remains limited.

Executives should treat this as a flow-control problem. AI increases production capacity faster than it increases the human capacity available to consume and evaluate the resulting output. More Slack messages, longer memos, larger reports, and frequent AI-generated code changes can lengthen review queues and increase cognitive demands across teams.

The first response is to reduce unnecessary output. Prompts should explicitly request concise answers. Meeting tools should extract deliverables, decisions, owners, deadlines, and status changes rather than requiring employees to read dozens of transcript pages. Teams should prioritize communications based on business importance instead of treating every AI-generated document as equally deserving of attention.

Software teams need similar discipline. An AI coding tool asked to resolve a specific bug may modify hundreds of unrelated lines. That expands the amount of code engineers must review and creates additional opportunities for defects. Prompts and development controls should restrict changes to the relevant function or file whenever possible.

Leaders should also discourage “workslop” from moving between employees. Sending pages of weak AI-generated material to a colleague transfers the effort of filtering and verification to that person. At organizational scale, this creates hidden work and weakens the productivity case for AI.

The executive metric should be useful work completed. Cycle time, error rates, rework, customer outcomes, and completed decisions can reveal whether AI creates meaningful gains. Raw message counts, document volume, and lines of generated code increasingly tell leaders very little about performance.

AI creates a cognitive cost that leaders need to manage

AI changes how people allocate attention during the working day. A software professional who previously spent long periods building and testing may now move repeatedly between prompting, reviewing, fact-checking, correcting, and re-prompting. Each transition requires a new decision and a new evaluation of machine-generated output.

Research involving 1,500 workers highlights this issue. A study by Boston Consulting Group and the University of California, Riverside, published in Harvard Business Review in March, used the term “AI brain fry” to describe the strain associated with intensive AI use. The researchers found that juggling multiple AI tools can cause decision fatigue and increase errors.

This matters because AI makes iteration extremely cheap. A user can request another draft, another solution, another rewrite, or another code implementation within seconds. Each additional option still requires human evaluation. The employee eventually becomes the limiting resource in the process.

The practical response starts with reducing constant interaction. Employees can group AI work into dedicated blocks rather than monitoring generated output throughout the day. Batch processing protects longer periods for focused work and reduces repeated transitions between creating, prompting, and reviewing.

Employees also need periods away from digital processing. A 15-minute walk, reading away from a screen, or another deliberate offline activity can provide recovery from sustained review and decision-making. This is a work-design issue because cognitive capacity directly affects accuracy and judgment.

Prompting itself can also become a source of wasted effort. Employees sometimes continue adjusting instructions in pursuit of a near-perfect result. Once an output is close enough, manually completing or refactoring the final portion can require less time and attention. Teams should give employees discretion to stop prompting when direct execution becomes more efficient.

Reusable prompts reduce another source of cognitive overhead. Employees can maintain two or three tested templates for recurring tasks, such as refactoring legacy code or drafting API documentation. These templates can already contain expected tone, constraints, validation rules, and output requirements. The employee then adjusts the relevant details instead of rebuilding instructions for every task.

For C-suite leaders, tool count also deserves scrutiny. Giving employees access to an expanding collection of AI products can create more decisions about which system to use, how to prompt it, and how to assess its output. Adoption should therefore be evaluated through workflow performance and quality, with attention to the cognitive demands created by each additional system.

The goal is sustainable productivity. AI can accelerate individual tasks while increasing the number of decisions surrounding those tasks. Organizations that manage attention, review workload, and interaction patterns will be better positioned to convert that speed into durable business performance.

AI is changing what employers look for in candidates

AI is changing both sides of hiring. Employers increasingly use applicant-tracking systems to filter resumes, while candidates use generative AI to prepare applications. At the same time, AI is changing job responsibilities themselves. Candidates now need to demonstrate measurable performance, sound judgment, adaptability, and effective AI use.

Leo Freitas, analyst and research manager at IDC Research, emphasizes evidence of performance. “You need demonstrable achievements,” he said. “It’s good to show highly human skills.” For candidates, that means showing what they accomplished and how their decisions created value.

Resume design is one practical constraint. Teresa Hill, founder and leader of growth marketing consultancy Anchor GTM, says traditional applicant-tracking systems rely heavily on structured information and keyword matching. Candidates should therefore mirror important terms from the job description. A resume for a “product marketing manager” position should use that complete phrase, even when the candidate also uses the abbreviation “PMM.”

Simple structure also improves machine readability. Conventional headings such as “Experience” and “Education” give parsers predictable information to process. Elaborate layouts and unusual section names can create unnecessary ambiguity in automated screening systems.

The content itself should center on outcomes. A statement that describes a responsibility gives an employer limited evidence of performance. A statement that connects the work to revenue, cost savings, conversion rates, delivery time, customer retention, reliability, or another relevant metric establishes a clearer record of business impact.

Generative AI can help candidates organize ideas and refine structure. Heavy reliance on generated language carries a different risk: applications can become generic and lose the candidate’s individual voice. Candidates should edit AI-assisted text until it accurately reflects their experience, decisions, and communication style.

The interview creates a more important test. Employers increasingly want to understand how candidates use AI. Strong answers explain which activities the candidate automates, which outputs receive additional scrutiny, and how the candidate verifies quality. Simply listing Claude, ChatGPT, Copilot, or other products demonstrates familiarity with tools. Explaining the decision process demonstrates professional judgment.

Hill also recommends emphasizing the capacity to learn rather than focusing heavily on expertise in a single AI product. This matters because AI products and capabilities change quickly. An employee who can learn new systems, evaluate their limits, and redesign workflows around them has a more durable capability.

Candidates should also present concrete examples of creativity, collaboration, problem-solving, and quality control. Technical professionals can strengthen these examples with measurable outcomes. They should be able to explain where AI was useful, where human review was required, and how they prevented errors from reaching customers or production systems.

For executives and HR leaders, the implication extends to recruiting design. Hiring processes should test judgment and demonstrated outcomes alongside technical knowledge. Interviews can ask candidates to review an AI-generated response, find problems, improve it, and explain the reasoning behind their decisions. This provides direct evidence of a capability that becomes more valuable as AI-generated work expands.

The hiring signal is shifting toward accountable execution. Employers need people who can use AI productively, detect weak output, understand business consequences, and remain responsible for the final result.

Career resilience requires AI skills, business knowledge, and continuous adaptation

Corporate investment is moving toward automation, and that shift is increasing job insecurity across the technology sector. The strongest career response is active adaptation. Employees should identify which parts of their jobs AI can perform, strengthen capabilities that remain valuable under automation, and use AI to raise their own performance.

Leo Freitas, analyst and research manager at IDC Research, expects substantial changes in work while rejecting the most extreme predictions of rapid, economy-wide displacement. “There’s this apocalyptical view that AI will take everyone’s job in a few years,” he said. “I don’t see that happening, but many things will change in the nature of how we work.”

Freitas recommends a direct self-assessment: determine whether AI can perform the work you currently do. That question can reveal where an employee’s future exposure is concentrated. Roles dominated by repeatable digital tasks face greater pressure as AI capabilities improve. Employees in those positions can move toward advanced responsibilities or areas such as security, where human oversight, accountability, and risk decisions remain important.

J.P. Gownder, vice president and principal analyst at Forrester Research, and Freitas both argue that fear of AI-driven layoffs currently exceeds the observed extent of job losses directly caused by AI. The underlying spending shift still matters. Companies are allocating more capital toward automation, which changes skill requirements, team structures, and expectations for individual productivity.

Continuous learning is therefore an operating requirement for a technology career. Employees can use courses, certifications, employer training, and practical projects to expand their skills. The objective should be transferable capability: understanding AI systems, evaluating their output, learning new tools quickly, and applying technology to important business problems.

Keith Jones, a cybersecurity researcher with more than 30 years of experience in the field, takes this approach. “I look at AI like it’s another new tool, and I’m going to learn it like I’ve learned any other tool,” he said. This keeps attention on practical capability as products and interfaces evolve.

Technical expertise alone also leaves value untapped. Employees who understand company strategy, competitive conditions, customers, internal processes, and commercial language can connect technical capabilities to business priorities. They can identify a problem, determine where technology can improve the outcome, explain the investment to management, and help deliver the change.

Institutional knowledge becomes especially useful in this environment. Employees who understand why internal systems work as they do, which constraints matter, and how different parts of the organization interact can apply that knowledge when designing or reviewing AI-enabled processes. Leaders should make this expertise visible and transferable rather than allowing it to remain isolated within individuals.

Employees can further strengthen their position by proposing new applications, customer-facing systems, process improvements, and technical capabilities that advance company goals. This shifts career development toward creating new business value as existing tasks become easier to automate.

For executives, workforce planning should follow the same logic. Training budgets should target capabilities linked to the company’s future operating model. Internal mobility can help employees move from highly automatable tasks into areas requiring deeper technical skills, customer context, security awareness, governance, or cross-functional decision-making.

AI adoption will continue to change roles even where headcount remains stable. Job descriptions will evolve, workflows will be redesigned, and expectations will rise. Organizations and employees that keep learning can absorb those changes faster and turn automation investment into stronger business performance.

Key executive takeaways

• AI shifts value toward judgment: AI can remove routine work and accelerate execution. Leaders should measure gains through quality, cycle time, business outcomes, and decision quality rather than output volume.
• Make human verification mandatory: AI can produce convincing errors, fabricated information, and unnecessary code changes. Set risk-based controls for reviewing, testing, and approving AI-generated work before it reaches production or customers.
• Control AI-driven workload creep: Upwork found that 77% of employees reported increased workloads from AI, while 88% of those with the highest productivity gains also experienced burnout. Reduce low-value output and prioritize concise, useful work.
• Manage the cognitive cost of AI: Constant prompting, checking, and switching between AI tools can increase decision fatigue and errors. Use batch workflows, reusable prompts, and focused work periods to protect employee attention.
• Hire for AI judgment and adaptability: Candidates increasingly need to demonstrate measurable outcomes, effective AI use, verification skills, and the ability to learn quickly. Hiring processes should directly test these capabilities.
• Build careers around durable business value: Employees should assess which tasks AI can automate, expand their skills, and connect technical expertise to business priorities. Leaders should support continuous learning and internal mobility toward higher-value roles.