Why productivity tools make knowledge workers busier, not better

Executive overview

New tools that speed up tasks or reduce cognitive effort consistently produce the opposite of their intended effect: more exhaustion, less output. Two mechanisms drive this. Faster task throughput pulls in more tasks behind it. Lower cognitive effort produces lower-quality work, requiring more total effort to reach a finished result.

The real reason we keep adopting tools that hurt us is pseudo-productivity — using visible busyness as a proxy for actual value created. Once you switch to measuring true output, the traps become obvious.

The two mechanisms that make tools backfire

• Speeding up a task increases its throughput — more tasks of that type flood in behind it
• Checking inboxes every two minutes (Microsoft's measured average) is the direct result of email making messages faster to send
• AI cuts time on administrative tasks, but queues of shallow tasks are effectively endless — the 94% rise in business tool use in the Avatrak study reflects this
• Reducing cognitive effort in the moment lowers output quality, requiring more back-and-forth to reach resolution
• Vague emails ("yeah, maybe?") save effort upfront but multiply the total email chain length
• AI-generated "work slop" — content that looks complete but lacks substance — creates more review and revision work than starting from scratch

Why we keep adopting tools that hurt us

• Pseudo-productivity: when true productivity is hard to measure, visible effort becomes the proxy
• Industrial-era managers could count units per worker-hour; knowledge work offers no equivalent number
• Workers internalised the same heuristic — busyness feels virtuous, idleness feels wrong
• Digital tools feed pseudo-productivity directly: higher task throughput looks productive; churning out AI-generated drafts looks productive
• Shifting to true productivity (value created per worker) reveals the traps that pseudo-productivity hides

Strategy 1: use a better scoreboard

• Identify the metric that actually moves the needle in your role — papers published, priority projects completed per month, user features shipped
• Track it before and after introducing a new tool
• If the number drops, the tool is a trap regardless of how productive it feels
• The same scoreboard distinguishes helpful uses of a tool from harmful ones — keep what helps, cut what doesn't
• The goal is the knowledge-work equivalent of Model T's produced per paid worker-hour

Strategy 2: identify and improve the true bottleneck

• The activities a tool speeds up are often not the constraint limiting your output
• Social scientists can use Claude Code to cut data analysis from three hours to 20 minutes — but if they produce one paper every two to three months, analysis time was never the bottleneck
• Academic tools (LaTeX, bibliography managers, digital communication) have improved many research steps without increasing per-researcher paper output — because the bottleneck was always something else
• In academic social science, the real bottleneck is negotiating access to interesting data sets, not processing speed
• In theoretical research, the bottleneck is deeply understanding enough prior work to see novel combinations
• Applying a tool to the actual bottleneck (email to expand data-access relationships) creates real gains; applying it elsewhere creates busyness

Strategy 3: separate deep from shallow work

• Protect scheduled time for high-concentration work before shallow tasks begin
• Side effects of productivity tools — inbox pressure, slop review loops, distraction — are contained to shallow hours and cannot infect deep work sessions
• Deep work sessions still use tools, but only those that directly advance the primary task
• This is a firewall, not a ban: the negative effects still occur, but they cannot crowd out the work that actually matters

On meetings: the same system dynamics apply

• Meetings multiply for rational organisational reasons: information exchange, risk distribution, participation signalling
• Framing this as individual bad behaviour ("people should just send an email") misses that meetings are a coordination system solving real problems
• The fix is replacing the system, not lecturing individuals
• Virtual meetings skyrocketed during the pandemic because zero-friction scheduling removed the social cost of convening — meeting counts stayed elevated even after office returns
• Reducing meeting volume requires solving the underlying coordination need another way:
  • Fewer concurrent projects per person reduces coordination overhead and paradoxically increases total throughput
  • Consolidated team check-ins (e.g. 45 minutes, three times a week) replace scattered ad hoc meetings
  • Office hours absorb five-minute questions that would otherwise become one-hour meetings
  • Fixed protocols for recurring work eliminate the need for unstructured coordination conversations
  • Higher barriers to calling meetings (Amazon-style detailed pre-read memos) reduce frivolous meetings and make the remaining ones more effective

On chatbots and psychological risk

• Brains anthropomorphise anything that produces fluent language — treating a chatbot as a social agent is automatic, not a choice
• For people prone to anxiety or rumination, chatbots actively extend and worsen those episodes: they never tire, always empathise, never push back
• Chatbots are also psychosis-amplifying machines — trained to be agreeable, they validate rather than challenge delusional beliefs
• Practical countermeasure: drop conversational register entirely; use terse search-query syntax ("tutorials Raspberry Pi Halloween decorations include links") — this reframes the tool as a search engine and reduces social anthropomorphisation