Visible work
The proposal, invoice, update or decision that appears in a task list. It is usually what gets counted.
FIELD NOTE · WORKFLOW RESEARCH
The invisible
workday.
Your calendar shows the meetings. It rarely shows the search, the re-reading, the “who has this?” and the careful restart after a ping.
Open the time ledger09:00Make43m
Proposal draft · the visible task. The useful unit is not “writing”; it is a checked proposal ready for a decision.
10:05Find17m
Searching for the latest pricing note, then checking whether the version is safe to reuse.
11:20Switch11m
A new message changes the priority. The unfinished draft remains active in the background.
13:40Rebuild23m
Reconstructing context after a handoff: what happened, what is current and what still needs an owner.
15:15Decide28m
Reviewing an exception and choosing whether to continue, ask, defer or stop.
Illustrative ledger, not a measured workday. Bar lengths are a visual scale only; they do not total an 09:00—17:00 day.
01 / The ledger
The day is full.
The work is scattered.
Most owners notice a delayed reply or a missed deadline. The system underneath is harder to see: information arrives in one place, context lives in another, and a person quietly becomes the connection between them.
Invisible work
Finding the source, checking the version, asking for missing information and remembering why the last decision was made.
Unowned work
The gap between “someone saw it” and “someone is responsible for the next safe step.”
THE QUESTION TO ASK
“Where does work wait for a person to remember what the system forgot?”
02 / The switching cost
Every restart
leaves a little
behind.
Research calls one part of this attention residue: after a switch, the previous task can keep occupying attention even when the new task is in front of you. Leroy’s experiments found that unfinished work made the transition harder; a clear stopping point and a deliberate handoff helped. [3]
Mark and colleagues observed a related tension in workplace interruptions. People may compensate by working faster after an interruption, but that speed can arrive with more stress. [2] Fragmented work also changes what “busy” means: the work surface can be full of motion while the important thread remains incomplete. [4]
That does not make every interruption bad. A customer escalation, a safety signal or a time-sensitive decision can deserve to cut the queue. The useful distinction is whether the interruption is chosen, visible and recoverable, or whether every notification is silently allowed to become a priority.
XT3 judgment
Protect attention before adding automation. A faster flow that creates more exceptions is a louder version of the same problem.
03 / The handoffs
Follow the thread
between people.
A workflow is not only the sequence of software screens. It is the sequence of decisions, permissions and small acts of interpretation that let one person safely continue another person’s work.
The request enters through email, form, chat or memory. Different doors mean different context.
Lost signalSomeone decides what it means, what is missing and how urgent it really is.
Hidden judgmentA person accepts the work, promises a next step or changes the source of truth.
Unclear ownerThe result is delivered, but the record of why and what changed may disappear.
Future reworkThe map is an XT3 model of a recurring business workflow, informed by research on fragmented work and interruptions. It is not a measured client journey.
04 / What to automate
Give the machine
the boring edges.
AI can be useful inside a well-defined flow. It should not be asked to invent the flow, the authority or the definition of done.
01First: remove looking.+
Make approved information findable. Normalize names, dates and status. If the business cannot agree which source is current, automation will only make uncertainty travel faster.
02Then: remove copying.+
Move structured facts between systems with ordinary rules. Use AI when the input is genuinely messy, and require a visible source and a review path for the result.
03Then: remove waiting.+
Route a complete request to the right owner, set a deadline and expose exceptions. An automated acknowledgment is not a completed outcome.
04Keep: judgment.+
Keep pricing, commitments, access changes and unusual customer situations with a person until the organization has evidence that a narrower action is safe.
The research is a boundary, not a promise. A controlled Copilot experiment found developers completed a particular timed task 55.8% faster with access to the tool. That is evidence that assistance can help in one setting; it does not prove that the surrounding workflow becomes faster. [6]
Find the expensive edge in your workflow ↗05 / A better day
Design for a clean
return.
The most useful automation leaves a person with a clear next step, not a new inbox.
Start by logging one week of a recurring workflow. Record what triggered it, where the person looked, where it waited, what had to be re-entered and what “done” meant. Do not optimize the story you wish were true.
Then choose one edge. Make the source visible. Give exceptions a queue. Keep the manual route beside the new one. Review the result with the people who receive the handoff, not only the person who configured it.
THE TESTCan the next person continue
without asking the last person
to reconstruct the day?
without asking the last person
to reconstruct the day?
1 / Observe2 / Name the outcome3 / Protect the exception4 / Test the return5 / Give it an owner
THE XT3 POINT OF VIEW
Make room for
good judgment.
You run the business. I handle the technology. Bring the workflow that keeps pulling people away from the work that matters; we will make the invisible cost visible and choose the smallest useful intervention.
Ian Dikhtiar · Hands-on fractional CTO · XT3
06 / Sources & methodology
A readable trail.
This report synthesizes public research available September 8, 2026. It does not report XT3 client measurements or claim that any particular automation will save a fixed amount of time.
The Microsoft number is platform telemetry, academic studies use particular populations and tasks, and the workflow maps and checklist are XT3 recommendations. Those kinds of evidence answer different questions and are not pooled into a single productivity score.
Use the ledger as a field instrument: observe a real workflow, record a baseline, test one bounded change and review the exceptions with the people who own the outcome.
- [1]2025 Work Trend Index: The year the Frontier Firm is born ↗
Microsoft · April 2025
Microsoft 365 telemetry and a survey of 31,000 knowledge workers in 31 markets. The report’s 275-interruptions figure is the 24-hour count for the top 20% by ping volume; the two-minute 9–5 framing is separate. It is not a universal count of all work interruptions. - [2]The Cost of Interrupted Work: More Speed and Stress ↗
Gloria Mark, Daniela Gudith & Ulrich Klocke · ACM CHI · April 2008
An observational and experimental study of workplace interruptions. It found that people may work faster after an interruption while experiencing greater stress; the setting and measures do not predict an individual business. - [3]Why is it so hard to do my work? The challenge of attention residue when switching between work tasks ↗
Sophie Leroy · Organizational Behavior and Human Decision Processes · July 2009
Two experiments on attention residue. Unfinished work can keep attention attached to the previous task; time pressure and a clear transition helped participants shift. Laboratory findings are a mechanism, not a productivity forecast. - [4]No Task Left Behind? Examining the Nature of Fragmented Work ↗
Gloria Mark, Victor M. Gonzalez & Justin Harris · ACM CHI · April 2005
Field study of fragmented work and task switching in an office environment. Used here for the distinction between a visible task and the hidden cost of repeatedly reassembling context. - [5]Work interruptions of office workers: The influence of the complexity of primary work tasks on the perception of interruptions ↗
Vera B. Rick, Christopher Brandl, Alexander Mertens & Verena Nitsch · Work · 2024 · first published July 2023
One-day diary study of 492 office workers in Germany. The study examines perceived interruption overload and task complexity; it is useful context, not a benchmark for all knowledge work. - [6]The Impact of AI on Developer Productivity: Evidence from GitHub Copilot ↗
Sida Peng, Eirini Kalliamvakou, Peter Cihon & Mert Demirer · Microsoft Research · February 2023
Controlled experiment on a timed JavaScript HTTP-server task. The reported speed result is bounded to the participants, task, tool and period; it does not establish that adding AI removes coordination work.