AI project planning helps turn a goal into a draft list of work, owners and dates. Imagine launching a customer page where people can see their invoices: before promising a launch date, you need to know what must be built, who can do it and who will approve it. AI can suggest a first plan, but the team must check every unknown. Below you can copy a brief, follow a worked example and review seven questions before approving a date.
Key takeaways
- Give the model scope, acceptance criteria, capacity and unknowns before asking for a schedule.
- Ask it to label assumptions and leave unsupported estimates open.
- Check the tasks that determine the finish date, the work time versus calendar time, and who is available.
- Approve a baseline only after owners have tested the plan against evidence.
What AI planning can and cannot do
A model can suggest a work breakdown, questions, dependencies and alternative sequences. It cannot know whether a supplier will meet an unpublished deadline, whether an expert is available or whether a sponsor accepts reduced scope. Treat each generated date or cost as a hypothesis until its inputs are checked. PMI provides AI learning resources for project professionals; the brief and checks below are Altirya's editorial method, not a PMI standard.
Separate effort (active work), duration (elapsed time, including waiting and handoffs) and capacity (time a suitable person can actually allocate). Six days of engineering effort cannot finish in one week if that engineer has only three project days available.
Copyable planning brief
Replace the example values before using this brief. Mark an unknown as “unknown” instead of letting a model fill it with a plausible invention.
| Input | Example to replace |
|---|---|
| Outcome and scope | Customers can view invoices and edit billing contacts; online payment is out of scope. |
| Acceptance | A finance owner verifies 10 sample invoices against the source ledger; contact changes pass an agreed permission test. |
| Target | Launch in 12 weeks is preferred, not contractually fixed. |
| Available roles | One engineer: 3 days/week; one designer: 1 day/week; product owner: 0.5 day/week. Finance reviewer unavailable in weeks 5–6. |
| Known dependencies | Existing login service and invoice API (the connection that supplies invoice data); confirm customer-account support and test-data access. |
| Open questions | Identity-provider customer model, API error handling and legal approval lead time are unknown. |
Request to copy: “Using this brief, propose deliverables and a plan for the next four weeks. For every executable task, show its output, owner role, effort range, dependency and acceptance evidence. Keep effort separate from elapsed duration. Mark unverified facts as assumptions, create discovery tasks for important unknowns and list decisions needed before any launch date can be committed. Do not invent staff availability or supplier promises.”
Worked example: a customer portal
This fictional portal is simply a page where customers sign in to see invoices. In this example, a technical spike is a short test to answer one technical question; acceptance means the customer-facing result has been checked; rollback means returning to the previous version if release fails.
The example is fictional. A first AI draft places authentication, invoice integration, testing and release into eight weeks. It looks complete but assumes the identity provider supports customer accounts, treats six engineering days as one calendar week and schedules finance acceptance when the reviewer is away.
| Draft problem | Review evidence | Plan correction |
|---|---|---|
| Authentication starts immediately. | Customer-account support is unconfirmed. | Run a two-day technical spike in week 1; decide by the end of week 2 whether to use the provider or revise scope. |
| Six days of integration fit in one week. | The engineer has three project days per week. | Allow at least two elapsed weeks after the spike, then check API and review waits separately. |
| Finance signs off in week 5. | The finance reviewer is unavailable in weeks 5–6. | Prepare test invoices earlier and move acceptance to week 7; name the finance owner. |
| Release follows coding. | Security review, support readiness and rollback criteria are missing. | Add review, acceptance and release gates before committing the launch date. |
The 12-week target remains a conditional target. The team cannot approve it until the identity decision, finance slot and security review are confirmed. If the identity spike fails in week 2, record a response in the risk register and replan scope or date. The capacity worksheet helps test the revised sequence against real availability.
Put it into practice. Want to test your brief? Create a first project in Altirya, then use the seven checks below before committing to a date.
Seven checks before approving an AI-generated plan
- Outcome: Can each major branch be traced to an accepted deliverable?
- Ownership: Does each executable task have a person or agreed owner role?
- Dependencies: Is each predecessor truly required, and is the needed output named?
- Estimates: Are effort ranges, elapsed duration, costs and their evidence distinct?
- Capacity: Can the required skills fit the proposed weeks after other commitments?
- Risk: Are important unknowns linked to a discovery task, trigger or response?
- Decision: Are acceptance, security, change and release approvals visible?
For each failed check, record the correction, owner and next decision date. A confident-looking plan is still a draft when its critical assumptions are unresolved.
Use AI planning responsibly
Use an approved tool and the organisation's data rules before entering customer records, personal information or confidential material. Have the people doing and accepting the work review the proposal. Record which estimates came from historical data, expert judgement or an AI suggestion. NIST's AI Risk Management Framework describes governance, mapping, measurement and management of AI risks; applying human review here is our practical inference from that broader framework.
To try this in Altirya, create a project from the brief, then inspect the suggested tasks one by one. Add a person responsible for each task, check the dates against actual availability and keep unanswered questions visible before you approve a deadline.
Frequently asked questions
Can AI create a reliable project plan?
AI can propose a useful first draft when the brief includes scope, acceptance criteria, dependencies and capacity. A team still needs to verify assumptions, estimates, owners and approvals before committing dates or costs.
What should an AI project planning brief contain?
State the outcome, included and excluded scope, acceptance evidence, preferred or fixed date, available roles, known dependencies and unresolved questions. Mark unknowns explicitly and follow your organisation’s rules for confidential data.
How do I check an AI-generated schedule?
Trace tasks to deliverables, test dependencies, separate effort from elapsed duration, compare required skills with actual weekly capacity, and name the people who approve acceptance and release.
