Part 2 · Working together

AI as a Planner: Breaking Down Big Work and Cutting Cloud Cost (Late 2025 to Early 2026)

As an architect, the hardest part of a project is often the first week, reading a long, complex requirement and turning it into a plan. From late 2025 I used AI to read with me, break big work into a work breakdown structure (WBS) of small tasks, and look for cloud cost savings. The method, and why the decisions stay with me.

Module: 06 of 11 Level: Part 2 · Working together Time: 7 min read Published:
All modules in this track (11)

A new piece of work lands. Fifty pages of requirements, three diagrams that don't quite agree, and a meeting on Monday where everyone expects a plan. In the past, the first few days went on reading, re-reading, and sticky notes. Now the first hour goes on a different kind of conversation.

From late 2025, I started using AI for the work before the work: understanding it, breaking it down, and deciding where the money goes.

Reading

Complex documents, made small

Long requirements and architecture documents are hard because everything is connected and nothing is in order. AI turned out to be a very good reading partner, as long as I ask the right things:

  • "Summarise this in ten bullet points, then list every assumption it makes."
  • "Where do these two sections disagree?" Contradictions are much easier to find with a second reader.
  • "What questions should I ask the authors before I start?" This one alone saves days.

The goal is never a summary to trust instead of the document. It's a map that tells me which parts of the document to read closely.

Planning

From one big task to many small ones

A work breakdown structure (WBS) splits a big piece of work into phases, then tasks, each small enough to estimate and finish. Building one used to take me a day of thinking. Now it's a conversation in five steps:

  1. Give it the requirement, plus the constraints: team, deadline, budget.
  2. Ask for phases: big stages, each with a clear outcome.
  3. Ask for tasks: small enough to finish in a day or two, with their dependencies.
  4. Ask what "done" means: an acceptance check for every task.
  5. Review it myself: reorder, cut, merge, and add what only I know.

The AI is fast at the first draft and good at spotting forgotten work: testing, documentation, access requests, cut-over, rollback. I'm the one who knows the team, the politics, and the history, so the final plan is mine.

Cost

A second pair of eyes on the bill

In 2026 I also started using AI for cloud cost optimization. The pattern is the same as debugging: AI suggests, data decides.

  1. Give it the facts. A cost breakdown by service, usage trends, and how the system is used.
  2. Ask for hypotheses. Idle or oversized resources, storage in the wrong tier, data transfer between zones, workloads that could use Savings Plans or Spot.
  3. Check each one with real data. Utilisation metrics, access patterns, and the actual bill.
  4. Decide with the people who own the system. A saving that breaks a recovery plan or a compliance rule isn't a saving.

The value isn't that AI knows the answers. It's that it suggests places to look that I might have skipped.

Try it in your workflow: plan your next piece of work with AI
  • Take the next requirement document you get. Ask for a ten-bullet summary, the assumptions, and the questions to ask the authors.
  • Ask for a WBS: phases, then tasks of one or two days, each with an acceptance check.
  • Ask: "What work is commonly forgotten in a project like this?" Add what applies.
  • For cost: export last month's cost by service, and ask for five hypotheses to check. Check them before changing anything.
What I'd tell you

If you're at this stage

  • Use AI to read, not to skip reading. Let it show you which parts deserve your full attention.
  • The first draft of a plan is cheap. The judgment is yours. Only you know the team, the risks, and the history.
  • Cost ideas are hypotheses. Check every one against real usage before you change anything.
What would you do?

Quick check

3 questions. Decide on your answer to each, then check it.

  1. Scenario

    AI turns a 50-page requirement into a neat 10-bullet summary. What's the best way to use it?

    1. Share the summary with the team instead of the document
    2. Use it as a map to find the sections to read closely, and check it against the document
    3. Ask the AI to estimate the project from the summary
    4. File the document, since the summary covers it
    Show answer

    Use it as a map to find the sections to read closely, and check it against the document

    A summary can drop the one detail that matters. Use it to decide where to look, not as a replacement for reading.

  2. Scenario

    An AI-generated WBS looks complete, but has no tasks for testing, access requests, or rollback. What does this tell you?

    1. You need to review it carefully and ask what's commonly forgotten, because the first draft is never the full plan
    2. The project doesn't need those tasks
    3. The AI is wrong, so you shouldn't use it for planning
    4. You should add 20% to every estimate instead
    Show answer

    You need to review it carefully and ask what's commonly forgotten, because the first draft is never the full plan

    First drafts of plans miss the unglamorous work. Asking directly for forgotten work, then reviewing with what only you know, makes the plan real.

  3. Scenario

    AI suggests moving a storage bucket to a cheaper tier to save money. What do you check first?

    1. How much money it saves
    2. How often the data is read, and whether any recovery or compliance rule depends on it
    3. Whether the AI has suggested this before
    4. Whether other teams have done the same
    Show answer

    How often the data is read, and whether any recovery or compliance rule depends on it

    Cheaper storage tiers can charge more for reading data back or be slower to restore. Usage and constraints decide whether the saving is real.

Next: Chapter 07. Late 2025, when agents started delivering platform code.

Where are you on your own AI journey?

If you're bringing AI into your own work, or you're stuck at one of these stages, I'd be glad to hear how it's going and compare notes on what's worked.

anupojuprudhvi@gmail.com