I Tracked Every Task Switch for Two Weeks - Then Automated the Worst Ones
I Tracked Every Task Switch for Two Weeks
For two weeks I logged every time I switched between applications or tasks - not just app switches, but actual context switches where I had to reload mental state. The average came to 47 context switches per day.
That number surprised me at first. Then I found the research that explains exactly why it hurt so much.
What the Research Says
Gloria Mark at UC Irvine found it takes an average of 23 minutes and 15 seconds to fully refocus after an interruption. A 2022 Harvard Business Review study found knowledge workers toggle between apps nearly 1,200 times per day and spend almost 4 hours per week just reorienting themselves after those switches. The American Psychological Association estimates chronic task-switching consumes up to 40% of productive time.
For developers specifically, the cost is even higher. Context reloading involves mental state that is expensive to reconstruct - the current state of a bug, which variable holds what, why a particular approach was chosen. Each switch does not just waste 23 minutes; it wastes the cognitive investment you made building up that specific context.
My Worst Offenders
After two weeks of logging, the patterns were clear. My most expensive switches were not interruptions - they were workflows that required touching three or four apps to complete one logical task.
Project status update. This required switching between a project management tool, Slack, a spreadsheet for tracking hours, and sometimes email for context. Each individual action took 30 seconds. The mental overhead of maintaining context across all four apps added 5-10 minutes of friction every single time.
Filing an expense report. Receipt in email, expense tool, bank statement in a browser tab, approval form. Four apps, one task. I ran this workflow roughly eight times per month. Each time felt like starting from zero.
Deploying a change. Build output in one terminal, deploy script in another, staging URL in a browser, Slack to announce to the team. The actions were mechanical but the switching cost was real.
The Pattern: Cross-App Workflows Are the Expensive Ones
The key insight from the data was that automating within a single app rarely helps much. The bottleneck is transitions between apps. When you are in Slack, you are in Slack. When you have to leave Slack, open a spreadsheet, update a cell, copy a value, go back to Slack, paste it, and check the project management tool - that is four context switches for one logical operation.
An agent that can operate across applications simultaneously - reading from one, writing to another, verifying in a third - eliminates the most expensive kind of context switch because it never has to switch at all. It holds all the state in its context window.
Setting Up Cross-App Automation
The workflows I automated had a common structure: trigger (some event or state), read (gather data from app A), transform (format or calculate something), write (update app B), verify (confirm the update landed). Once I could describe workflows in that structure, I could delegate them.
The project status update became: read current tickets from the project tool, read recent Slack messages about those tickets, update the spreadsheet, post a summary to the standup channel. The agent runs this in under two minutes with no app-switching from me.
The expense report became: read the receipt details from email, extract the amount and vendor, fill the expense tool fields, attach the email as documentation. One command, no mental overhead.
Results After Six Weeks
Daily context switches dropped from 47 to around 18. The 18 that remain are the ones that genuinely need my judgment - creative decisions, ambiguous situations, work that requires understanding nuance.
The time saved is real but secondary. The more significant change is the reduction in cognitive residue. When you context-switch 47 times a day, you carry fragments of every switched-away task with you. At 18 switches, I finish the day with a cleaner mental state.
What Still Needs Human Attention
Automating cross-app mechanics works well for workflows with a clear structure and predictable inputs. It does not work well for creative work, decisions that require judgment about tradeoffs, or anything where the inputs are ambiguous.
The right split is: the agent handles the mechanical cross-app shuffling; I handle the parts that need human thought. Getting that split right is the actual work - and it took more iteration than I expected to identify which tasks fell into which category.
Fazm is an open source macOS AI agent. Open source on GitHub.