Personalized Learning

Most productivity tools treat every user the same way. They have fixed algorithms, preset categories, and one-size-fits-all suggestions. The problem is, everyone works differently.

Some people thrive on detailed task lists with 15-minute increments. Others freeze up when they see too many small tasks—they prefer big chunks of focused work. Some consistently underestimate how long things take; others pad everything with extra time "just in case."

GDT doesn't try to change how you work. It learns how you work.

The Problem with Generic Estimates

When you ask any task management tool to estimate time, it applies some universal formula. "Writing" tasks get X hours. "Coding" tasks get Y hours. But these estimates are almost always wrong for you specifically.

Here's what actually happens: You estimate a report will take 2 hours. It takes 5. Next time, you estimate 2 hours again—because that's the "reasonable" estimate. It takes 5 hours again. This cycle repeats indefinitely.

The issue isn't that you're bad at estimating. The issue is that the tool has no memory of your actual patterns. It can't learn that you specifically take 2.5x longer on writing tasks than the "average" estimate suggests.

GDT breaks this cycle. When you complete a task, it compares your estimated time against actual time. Over multiple completions, it builds a picture of your personal estimation bias. The next time you plan a similar task, GDT adjusts its suggestions based on your history—not some generic formula.

Task Granularity Is Personal

Ask ten people how they want their tasks broken down, and you'll get ten different answers.

Person A wants everything in 15-minute chunks. "Review document section 1" as a separate task from "Review document section 2." This level of detail helps them maintain momentum—they get the satisfaction of checking off boxes frequently.

Person B finds this maddening. Too many tasks creates anxiety. They prefer "Review entire document" as one task, then they'll naturally break it down in their head as they work.

Neither approach is wrong. They're just different cognitive styles.

Traditional tools force you to adapt to their granularity. GDT adapts to yours. When you say "this is too detailed" or "break it down more," GDT remembers. Future decompositions automatically adjust to match your preference.

This isn't a setting you configure once. It's a continuous learning process. As your needs change—maybe you want more detail for unfamiliar projects, less for routine work—GDT's understanding evolves with you.

Context That Persists

Every conversation with a traditional AI assistant starts from zero. You explain your project, your constraints, your preferences—then the session ends and it's all forgotten. Next time, you start over.

This isn't how human collaboration works. When you work with a colleague over months, they build understanding. They know which projects you're stressed about. They remember that you prefer morning meetings. They've learned not to schedule deep work right after lunch.

GDT builds this kind of persistent context. When you mention a project in conversation, it becomes part of your working memory. When you frequently use certain tags or work on specific projects, GDT notices. This context informs future suggestions without you having to repeat yourself.

For example, if you've been discussing your "Q1 Launch" project all week, and you say "add a task to review the marketing deck," GDT knows which project you mean. You don't need to specify "project:q1-launch" every time—the context carries forward.

The goal isn't to remember everything—that would be overwhelming. It's to remember what's relevant, so conversations can build on previous understanding rather than starting fresh each time.

How Learning Happens

GDT learns from three sources:

Your actions. When you create tasks, complete tasks, or organize your work, GDT observes patterns. Which projects come up most? Which tags do you use? How long do different types of tasks actually take?

Say you create 15 tasks this month, and 12 of them are tagged "backend" in the "platform" project. Next time you create a task, GDT will suggest these as defaults—because that's what you actually use.

Your feedback. When you tell GDT "this breakdown is too detailed" or "I underestimated that task," it's explicit learning. Your words directly shape future behavior.

You: That breakdown was too granular, I prefer bigger chunks
GDT: Got it. I'll use coarser granularity for future decompositions.

From then on, task breakdowns arrive at the level you prefer.

Your patterns over time. Some insights emerge only from accumulated data. Maybe you're more productive on certain days. Maybe tasks in a specific project always take longer than estimated. These patterns become clear through observation, not explicit feedback.

All of this happens locally. Your learning data stays on your machine in ~/.gdt/memory/. Nothing is sent externally. The personalization is truly yours.

What This Means in Practice

After a few weeks of using GDT, the experience changes noticeably.

Time estimates become more accurate. Instead of generic suggestions, you get estimates that reflect your actual pace:

You: Break down "write the API documentation"

GDT: Based on your history with documentation tasks (which typically
     take you 1.8x longer than initial estimates), here's a realistic plan:

     1. Outline structure (1 hour)
     2. Write endpoint descriptions (3 hours)
     3. Add examples and edge cases (2 hours)
     4. Review and polish (1 hour)

     Total: ~7 hours (adjusted for your pace)

Task decomposition matches your style. If you've consistently asked for coarser breakdowns, new decompositions arrive at that level by default. No need to adjust every time.

Projects and tags surface intelligently. When creating a task, GDT suggests the projects and tags you actually use:

You: Add a task to fix the login bug

GDT: Created task "Fix login bug"
     Suggested: project:platform, +backend, +urgent
     (Based on your recent work patterns)

     Apply these? [Y/n]

The tool starts feeling less like software and more like working with someone who knows your patterns.

The Broader Vision

Personalized learning is central to what makes GDT different from traditional productivity tools.

Most tools optimize for the "average user." They implement features that work reasonably well for most people. But you're not average—nobody is. The features that work for the average user might be exactly wrong for your specific needs.

LLM-native tools can take a different approach. Instead of hardcoding behavior, they can observe and adapt. Instead of forcing you into predetermined workflows, they can learn your natural patterns and support them.

This is what "the tool adapts to you" really means. Not just a conversational interface, but genuine personalization that emerges from understanding how you specifically work.

GDT is still early in this journey. The learning capabilities will deepen over time. But the foundation is here: a tool that gets better at helping you the more you use it.

Next Steps

• Task Decomposition — See how learning affects breakdown suggestions
• Privacy Controls — Understand how learning data is stored and managed
• Conversation Guide — Tips for giving effective feedback that shapes learning