Day 80

Day 80 - July 20, 2026: AI Built the Structure, Fluency Finished the Work

A Day 80 reflection on using Notion AI to organize a QA testing reference, then relying on manual editing, keyboard fluency, and judgment to finish the details.

Notion AI could assemble a useful QA testing reference page.

It could gather external resources, organize links, explain several kinds of software testing, and build a table comparing when one form of testing might be more appropriate than another.

It could not make a hyperlink blue.

That contrast was the small, funny center of Day 80.

The substantial part of the task went well. AI accelerated the research and gave the page a clear first structure. The stubborn part was a tiny visual detail that would normally seem easier than collecting and organizing all that information.

After asking several times for the traditional blue hyperlink styling I wanted, I stopped trying to improve the prompt. I highlighted the link text and changed its color manually.

The page needed both kinds of work.

AI created most of the structure. Fluency with the underlying tool finished it.

A First Draft That Was Actually Useful

The goal was not to test Notion AI for its own sake. I was learning Notion by building pages, and one of those pages became a reference for QA testing.

I asked the AI to find and organize useful external resources, include links, and explain different forms of software testing. This is exactly the kind of open-ended documentation task where a blank page can create more friction than the writing itself.

Notion AI removed much of that friction.

It produced a useful first draft with sections rather than a loose collection of notes. It grouped related ideas and presented the material in a form that I could review. It also used structured elements such as tables. One table compared test types and explained the situations in which one approach might be more suitable than another.

That organization mattered more than any individual sentence.

A QA reference page should help someone find distinctions quickly. A long paragraph about testing can be accurate and still be difficult to use. A comparison table makes the tradeoffs visible. Links turn explanations into starting points for deeper research. Headings provide a map of the subject.

The AI did not merely add words to the page. It helped establish that usable shape.

Of course, a generated reference is still a draft. External resources need judgment. Explanations need review. A confident structure does not guarantee that every distinction is complete or that every source is the best one.

But the draft was good enough to change the nature of the work. Instead of inventing the page from nothing, I could inspect, edit, and improve an existing artifact.

That is meaningful acceleration.

The formatting problem was modest but persistent.

Notion’s links can appear visually similar to the surrounding text. I prefer the familiar web convention of blue hyperlink text because it makes the clickable parts of a reference page easier to recognize at a glance.

I asked the AI to change the links, or their text, to that blue styling.

It did not make the change correctly.

I tried more than once because the request seemed so small and exact. The AI had just organized a substantial page about software testing. Surely changing the appearance of a link should be the easy part.

It was not.

The irony made the limitation more memorable. The AI could reason at the level of topics, categories, comparisons, and resources. It struggled at the level of one precise interface action.

This was not a reason to discard the useful draft. It was a reason to identify the boundary of the assistance.

The workaround took very little time. I selected the hyperlink text and changed its color myself. The important shift was recognizing that another prompt was no longer the fastest path.

Prompting has a cost, even when the individual request feels free. Each retry asks for attention, waiting, inspection, and another decision about whether the result is correct. When a direct edit is obvious and reversible, using the tool can be more efficient than continuing to negotiate with the AI.

The best workflow was not AI or manual editing.

It was AI for the broad structure and manual control for the exact finish.

A Shortcut Changed the Editing Experience

The manual fix exposed another small source of friction.

Selecting hyperlink text with the mouse was awkward because dragging over the link could open it instead of selecting it. A simple edit became a careful attempt not to trigger the thing I was trying to change.

I learned a keyboard-assisted way to select the text more reliably.

The exact key combination matters less here than what learning it changed. The link-color workaround was already easy in theory. The shortcut made it easy in practice.

That difference is part of tool fluency.

Knowing that an interface can perform an action is not the same as being able to perform it without breaking concentration. Repeated mouse corrections, accidental clicks, and tiny targeting problems create a tax that is almost invisible when each instance is considered alone.

A shortcut removes part of that tax.

It also changes the calculation around AI assistance. Once the manual action is fast and dependable, there is less reason to spend several turns asking an AI to reproduce it indirectly. Better command of the tool makes it easier to choose the right boundary between delegation and direct control.

The shortcut did not replace the AI-generated structure. It made the handoff from AI to human smoother.

Productivity Is Often Friction Removal

Later, I watched YouTube videos about how developers optimize their Mac setups and make everyday work more efficient.

It was encouraging to recognize tools and practices I had already adopted, including Visual Studio Code and Homebrew. That recognition mattered because workflow-improvement content can sometimes make productivity look like a collection of unfamiliar applications that must all be installed before serious work can begin.

My setup already had a useful foundation.

The videos also made the remaining opportunity clear. I can learn more keyboard shortcuts and evaluate applications that reduce repeated manual actions. The goal is not to collect software. It is to notice where ordinary work repeatedly slows down and decide whether a tool, shortcut, or better habit can remove that friction.

The hyperlink experience was a small example of exactly that process.

The frustrating action was not writing the QA reference. It was selecting a link without opening it. The improvement was not a new platform or an elaborate automation. It was a more fluent way to perform one recurring interaction.

Developer productivity is often discussed through major choices: editor, terminal, package manager, operating system, or AI assistant. Those choices matter, but much of the day-to-day experience is shaped by smaller loops:

Removing a few seconds from one loop seems trivial. Removing hesitation from a loop repeated throughout the day can change how the whole environment feels.

AI Assistance Still Requires Tool Knowledge

The Notion page and the Mac workflow videos pointed to the same lesson from different directions.

AI can reduce the effort required to begin. It can collect material, suggest a structure, create comparisons, and turn an empty page into something worth reviewing.

It does not eliminate the need to understand the environment where that work lands.

Someone still has to notice that the links are difficult to distinguish. Someone has to decide that blue text is the preferred convention for this page. Someone has to recognize when repeated prompting is producing less value than a manual edit. Someone has to know, or learn, how to make that edit without accidentally opening the link.

Those are not failures of automation. They are the parts of the workflow where judgment and fluency remain active.

The more capable the AI becomes, the more important that judgment may be. Strong first drafts can create a temptation to treat the remaining details as unimportant or assume that the system which produced the structure should also be able to finish every interface-level operation.

Day 80 offered a more practical model.

Use AI where it creates leverage. Review what it produces. Learn the tool well enough to recognize the remaining gap. Then finish the work through the shortest reliable path.

Sometimes that path is another prompt.

Sometimes it is selecting the text and making it blue.

A Learning Day, Not a Transformation Story

This was primarily a day of learning, documentation, and workflow improvement.

I did not emerge with a perfectly optimized Mac setup or a complete theory of AI-assisted knowledge management. I built Notion pages, created a useful QA reference draft with AI assistance, solved a small formatting problem manually, learned a more comfortable editing technique, and looked for ideas that could make future development work smoother.

That was enough.

Small workflow improvements are valuable precisely because they do not require a dramatic transformation. They accumulate through repeated use. A reference page becomes easier to scan. A link becomes easier to edit. A shortcut removes one interruption. Familiar tools become part of a more deliberate system.

The page itself captured information about testing.

The process of building it taught a different kind of QA lesson: evaluate the whole experience, not only whether the main function succeeded.

The AI successfully generated the content and structure. The result still needed a human to notice one awkward visual detail and finish it.

Both facts can be true.

Outcome

Day 80 produced a structured QA testing reference page in Notion and a clearer model for combining AI assistance with direct tool use.

Notion AI gathered and organized external resources, explained multiple forms of software testing, included links, and created structured comparisons such as a table describing when different testing approaches may be appropriate. The result was a useful first draft that still required review and judgment.

The AI did not reliably apply the preferred blue styling to hyperlink text despite repeated requests. A manual color change completed the formatting. A keyboard-assisted selection technique made that edit less frustrating than dragging over link text with the mouse and risking an accidental open.

Public videos about developer Mac workflows reinforced the same theme. Visual Studio Code and Homebrew were already part of my setup, while keyboard fluency and carefully chosen friction-reducing applications remained areas for improvement.

The central lesson was not that AI failed because one link stayed the wrong color. AI handled the broad research and organization well. The lesson was that effective work came from combining that leverage with tool knowledge, manual control, and judgment about when to stop prompting.

AI built the structure.

Fluency finished the work.

Definition Of Done

Day 80 reached the AI-assisted documentation and workflow-learning checkpoint: