In the previous posts, I looked at Chat, Ask, Edit, and Agent, and how each supports a different moment in AL development.
- Introduction to GitHub Copilot in VSCode for AL Development – think about IT
- Chat, Ask, or Edit? Using GitHub Copilot Effectively in AL Development – think about IT
- Copilot Agent in AL: Using It Well Without Losing Control – think about IT
All of them do something:
-
Chat helps you think
-
Ask helps you understand
-
Edit helps you improve
-
Agent helps you implement
Plan mode is different.
Plan does not write code.
Plan does not change files.
Plan deliberately slows things down.
And that is precisely why it matters.
Why Plan Mode Exists
Copilot is very good at acting quickly.
Sometimes too quickly.
In AL projects, many issues are not caused by incorrect code, but by unclear intent:
-
responsibilities are not well defined
-
business logic drifts into the wrong layer
-
objects are created “just in case”
Plan mode exists to address that problem.
It creates a moment where:
-
nothing is executed
-
assumptions are surfaced
-
intent is made explicit
Before any code is written.
The Problem Plan Mode Solves in AL Projects
AL is an opinionated environment.
Where code lives matters:
-
validation belongs in tables
-
business logic belongs in codeunits
-
pages should remain thin
When intent is not explicit, even experienced developers can drift.
When Copilot is involved, that drift simply becomes visible faster.
Plan mode helps by forcing a simple but often skipped question:
What are we actually trying to build — and why?
What Plan Mode Actually Produces
It is important to set expectations correctly.
Plan mode:
-
does not generate AL code
-
does not modify files
-
does not create objects
Instead, it produces:
-
a structured approach
-
a sequence of steps
-
a division of responsibilities
-
explicit assumptions and trade-offs
Think of Plan mode as a design conversation, not a coding tool.
Agent acts.
Plan proposes.
A Small AL Scenario Where Plan Makes Sense
Consider a familiar request:
“We need to add a Loyalty feature for customers.”
Even this simple request raises questions:
-
Is this just a field, or something more?
-
Where should calculation logic live?
-
Will this evolve over time?
-
Do we need configuration or setup?
Jumping straight to Agent is tempting.
But without clarity, Agent will start making decisions for you.
Plan mode is the moment where you pause and decide with Copilot instead.
Using Plan Mode in Practice
When using Plan mode, I do not ask for code.
I ask for structure.
Typical questions I ask in Plan mode include:
-
Which AL objects are involved?
-
Where should responsibilities live?
-
What should exist now, and what can wait?
-
How might this evolve later?
The goal is not to design everything upfront.
The goal is alignment.
📸 Screenshot — Copilot Plan mode selected
Using Copilot Plan mode to reason about structure before writing AL code.
How Plan Mode Improves Agent Results
Plan mode becomes most valuable when combined with Agent.
Instead of starting Agent with:
“Build a Loyalty feature”
You can start with:
“Implement step one of the agreed plan.”
That change alone makes Agent:
-
more predictable
-
more focused
-
easier to guide incrementally
Plan does not slow development down.
It reduces rework.
When Plan Mode Is Worth the Time
Plan mode is especially useful when:
-
a feature touches multiple objects
-
architecture matters
-
Agent will be used extensively
Plan mode is usually unnecessary when:
-
making a small refactor
-
adjusting a single trigger
-
fixing a local issue
Like all Copilot modes, Plan is situational.
It is not something you need all the time.
My Typical Workflow: Plan → Agent → Edit
In practice, my Copilot workflow often looks like this:
-
Plan to clarify intent and structure
-
Agent to implement incrementally
-
Edit to refine code
-
Ask to review or explain decisions
Each mode has a clear responsibility.
Skipping Plan does not make Copilot ineffective.
Using Plan makes the rest far more predictable.
Why Plan Mode Is Easy to Skip
Plan mode is easy to ignore because:
-
it feels slower
-
it does not produce code
-
it looks like “extra work”
But in AL projects, upfront thinking is rarely wasted.
Plan mode externalises that thinking and makes it explicit — both for you and for Copilot.
📸 Screenshot — Plan output overview
Using Plan mode to agree on responsibilities and sequencing before execution.
📸 Screenshot — Refining intent in Plan mode
Refining architectural intent in Plan mode before implementation.
Closing Thoughts
Plan mode is not about AI.
It is about intent.
Copilot works best when intent is clear, boundaries are explicit, and responsibilities are agreed upon before execution starts.
If Agent is about doing,
Plan is about deciding.
And in AL development, that distinction matters more than it might seem.
Discover more from think about IT
Subscribe to get the latest posts sent to your email.