The Claude AI Planning Framework: A System for Knowledge Workers

Most planning frameworks fail for one of two reasons: they’re designed for a world without AI, or they’re designed for AI tools without accounting for how specific tools actually work.

This framework is built for Claude specifically. It accounts for what Claude does well (long-context reasoning, structured output, honest pushback), what it doesn’t do (proactive reminders, memory outside Projects, live data access), and how to sequence planning conversations across a week so the system compounds rather than just repeats.

We call it the Claude Planning Stack — three infrastructure layers and five conversation types that together form a complete planning practice.

Layer 1: The Project Foundation

Every effective Claude planning system starts with a configured Project. Without it, each conversation starts without context. With it, every conversation starts knowing your role, your goals, and your constraints.

A well-configured planning Project has three components:

System instructions — your role, current goals, standing constraints, and planning preferences. Written once, updated quarterly. This is the single most important configuration step.

Reference documents — anything Claude should know that isn’t captured in instructions. OKRs, project briefs, sprint backlogs. Keep these focused: three tight documents outperform one sprawling one.

Conversation discipline — the habit of opening a fresh conversation inside the Project for each planning session, rather than appending to an existing long thread. Fresh conversations reduce noise while the Project instructions maintain continuity.

The Project is infrastructure. The conversations are where the work happens.

Layer 2: The Five Conversation Types

The framework has five typed conversations. Each has a specific purpose, a specific trigger, and a specific prompt structure.

Type 1: The Weekly Plan

When: Sunday evening or Monday morning, once per week. Purpose: Translate your goals, commitments, and constraints into a realistic week. Output: A formatted Artifact — one row per day, columns for Focus Block, Priority Tasks, Admin/Comms.

Prompt:

It's [day, date]. Weekly plan time.

Calendar this week: [paste events with durations]
Key deadlines: [list]
Carry-over from last week: [list]
Energy forecast: [describe when you're sharpest and any constraints like travel or illness]

Build a weekly plan as an Artifact. One row per day. Columns: Focus Block | Priority Tasks | Admin/Comms | Flags.

In the Flags column, note: overloaded days, dependency conflicts, and anything I'll likely push that I shouldn't.

The “Flags” column is what separates this from a basic schedule. It makes Claude’s honest assessment visible without requiring you to ask for it separately.

Type 2: The Daily Check-in

When: Each morning, 5 minutes. Purpose: Confirm or adjust today’s plan given what’s changed overnight. Output: A clear, 3-item priority order with honest caveats.

Prompt:

Daily check-in. [Date].

Today's plan from my weekly plan was: [paste row from Artifact]
What's changed: [new fires, reschedules, energy drop, etc.]
Energy right now: [1–10]

Given this, what should I actually do today? Give me three priorities in order, and tell me the one thing I'm most likely to avoid that I shouldn't.

The final question — “the one thing I’m most likely to avoid” — forces Claude to reason about behavior, not just task sequencing. That’s where planning conversations get genuinely useful.

Type 3: The Project Decomposition

When: When starting a new project, or when a project has stalled. Purpose: Break a complex goal into sequenced milestones with dependencies and decision points. Output: A milestone table with effort estimates and the most important decision at each stage.

Prompt:

I need to complete [project name] by [deadline].

Here's what I know: [2–4 sentences describing the project]
Known constraints: [time budget, dependencies on others, skill gaps]
What I'm uncertain about: [the parts that feel unclear]

Break this into 4–6 milestones. For each: estimated effort, what it depends on, and the single most important decision I need to make before I can start it. Format as an Artifact table.

The “most important decision” column is the key. It forces both you and Claude to identify where the real work is — not the doing, but the deciding.

Type 4: The Priority Triage

When: When you have more competing tasks than available time. Purpose: Get a ranked order from an overwhelmed task list using explicit reasoning. Output: Ranked list with Claude’s reasoning on the two hardest calls.

Prompt:

I have too much to do and limited time. Here's everything competing for my attention today:
[list all tasks]

My constraints: [hours available, energy level, any hard deadlines]
Stakeholder pressures I'm managing: [brief — who is waiting on what]

Rank these by: urgency × importance × only-I-can-do-this. 

Then explain the two hardest ranking calls you made and why.

Explicitly requesting the “two hardest calls” is a technique worth making habitual. It surfaces the reasoning behind the ranking, which lets you calibrate against your own judgment rather than just accepting the output.

Type 5: The Weekly Retrospective

When: Friday afternoon, once per week. Purpose: Diagnose the difference between your plan and your week — and find the pattern. Output: A short diagnosis with one structural recommendation for next week’s planning.

Prompt:

End of week review. [Date].

I planned to: [paste Monday's weekly plan Artifact]
What actually happened: [brief honest account]
Wins: [list]
Pushed items (things I planned but didn't do): [list]
Unexpected items that took time: [list]

Diagnose my planning this week. Don't just say "be more realistic." Tell me:
1. Where specifically my estimates were off
2. Whether the pattern suggests a systemic issue (e.g., I always underestimate meeting prep)
3. One structural change to make to next week's plan setup

The instruction to “not just say ‘be more realistic’” is deliberate. Without it, retrospective conversations tend toward vague encouragement. With it, they produce specific, actionable diagnosis.

Layer 3: MCP Integration (When You’re Ready)

Once the Project-and-conversations rhythm is established, MCP can extend the framework significantly.

MCP (Model Context Protocol) allows Claude to connect to external tools. For planners, the most valuable integrations are:

• Calendar access: Claude can read what’s actually on your calendar rather than relying on what you paste in
• Task manager access: Pull your open tasks directly into the triage conversation
• Notes or knowledge base: Surface relevant past decisions or meeting notes without manual copying

MCP setup requires a compatible server for each integration. The technical setup is beyond this article’s scope, but for a full walkthrough of what’s possible, the complete guide to planning with Claude AI covers it in detail.

The important caveat: MCP adds complexity. If you’re new to the framework, get the Project-and-conversations layer working first. MCP is a multiplier on a working system, not a substitute for one.

How the Five Conversations Compound

The power of the framework isn’t in any single conversation — it’s in the sequence.

The Weekly Plan sets the direction. The Daily Check-ins keep you aligned as reality diverges from the plan. The Retrospective creates the feedback loop. Project Decompositions prevent large work from being underprepared. Priority Triaages handle the firefighting that every week eventually involves.

Each conversation references or builds on the others. The Retrospective informs next week’s Weekly Plan. The Daily Check-in references the Weekly Plan. The Project Decomposition feeds into the Weekly Plan’s Focus Block assignments.

Over four to six weeks, Claude also accumulates a form of operational memory through your Project instructions — because you update those instructions based on what you learn in Retrospectives. The system compounds.

What to Put in Project Instructions After the First Month

After your first month, you’ll have identified patterns. Update your Project instructions with what you’ve learned:

Calibration notes from experience:
- I consistently underestimate meeting prep by 30–40 minutes
- My focus blocks on Monday mornings are rarely protected — schedule deep work Tuesday/Wednesday instead
- I push research tasks. If I list research as a priority, ask me why it's not delegated or dropped.
- My energy on Fridays is low. Reserve admin and communication for Friday afternoon.

This is the compounding element. Claude can’t learn your patterns autonomously — but when you encode what you’ve learned into instructions, every future conversation benefits.

For knowledge workers who want this process tightly integrated with a calendar app that already understands how planning context works, Beyond Time is worth exploring — it’s designed around the same principle that the plan and the schedule need to stay in sync.

What the Framework Is Not

This framework does not replace a task manager. It reasons about what to do, but it doesn’t track completions, surface due dates automatically, or push reminders.

It does not handle team coordination. If you need to coordinate with others, the conversations described here inform your planning, but the coordination work happens in your project management tool.

It is not an always-on assistant. Claude doesn’t proactively check in with you. You run these conversations on a schedule. If you skip a week, you’re missing the feedback loop — not just a single session.

Those limitations aren’t failures of the framework. They’re the honest scope boundary. Within that scope — helping a single knowledge worker reason about their time, priorities, and projects — the framework is more useful than most planning approaches that don’t use AI at all.

Your First Week with the Framework

Run the conversations in this order:

1. Today: Set up the Project and write your instructions (15 minutes)
2. This Sunday or Monday: Run your first Weekly Plan conversation
3. Each morning this week: Run the Daily Check-in (5 minutes each)
4. This Friday: Run the Weekly Retrospective and update your Project instructions based on what you learn

By the end of the week, you’ll have a calibrated Project and a clear sense of where the framework fits your actual workflow. Adjust from there.

Related: The Complete Guide to Planning with Claude AI · How to Plan with Claude AI Step by Step · What Claude Does Well for Planning · Complete Guide: Setting Goals with AI

Tags: Claude AI planning framework, planning system, Claude Projects, knowledge work planning, AI productivity framework