Dependencies: sequencing work and spotting blockers

By the end, you can

• Explain what a dependency is, and why an unsequenced task list can hide one (PM-2).
• Use AI to draft a first-pass dependency map from a task breakdown, and explain why a named owner must confirm which dependencies are real before relying on the sequence (PM-2).
• Identify a hidden or missing dependency in a sample task list that an AI-only pass overlooked (PM-1, PM-2).

Before you start

This is Module 2, Lesson 2 of the AI-Powered Project Management course. It builds on Lesson 1's milestone list and sliced task breakdown, and turns that list of tasks into a sequence. It assumes you have completed Lesson 1 and have used at least one AI chat assistant for drafting text.

A dependency is a relationship, not a guess

A sliced task breakdown is a list. A list has no order until something tells you which tasks must happen before which others. APM's project management glossary defines a dependency plainly: "a relationship between activities in a network diagram." A network diagram, the same glossary explains, is "a model of activities and their dependencies used in scheduling." Put together: a dependency is not a preference about what would be tidy to do first, it is a genuine relationship — one task cannot properly start, or cannot properly finish, until another one does.

Some dependencies are about order: mentor orientation cannot run before mentors are recruited. Some are about a shared, limited resource: two tasks that both need the same one person cannot both happen at once, however unrelated they otherwise are. Missing either kind has the same effect — a plan that looks achievable on a task-by-task basis but cannot actually run in the order assumed. The glossary's definition of a critical path shows why this compounds: "a sequence of activities through a precedence network from start to finish, the sum of whose durations determines the overall duration." The longest chain of genuinely dependent tasks — not the busiest person, not the most tasks — is what actually sets how soon the project can finish. Miss one real dependency on that chain, and the whole plan's finish date is wrong, not just one task's.

Why AI-surfaced dependencies still need a human check

Given a task breakdown, an AI assistant can propose a plausible sequence: it can read "recruit mentor pool" and "run mentor orientation session" and correctly guess that recruiting comes first. That is a genuine service — sequencing a long task list by hand is tedious and easy to get wrong. It is not the same as knowing your project's real blockers.

The Open Worldwide Application Security Project's (OWASP) generative-AI security guidance names the risk directly: "overreliance occurs when users place excessive trust in LLM-generated content, failing to verify its accuracy." An AI assistant reading a task list has no access to a vendor's actual lead time, a colleague's holiday, a shared piece of equipment, or a compliance sign-off that has to happen before work can proceed — the kinds of dependency that come from your organisation's real constraints, not from the words on the task list. It can guess at ordering from task names; it cannot know what you have not told it. The US National Institute of Standards and Technology's (NIST) AI Risk Management Framework describes an AI system as something that "can, for a given set of objectives, generate outputs such as predictions, recommendations, or decisions" — a proposed dependency map is exactly that kind of generated recommendation, and nothing in that description includes the system confirming which proposed dependencies are actually real. That confirmation belongs to the named owner, who checks each one against the organisation's real constraints before the sequence is relied on.

The response also matters once a real blocker turns up. If a genuine dependency forces a schedule change, a workaround, or reordering tasks against the AI's original sequence, that is itself a consequential call, not a drafting exercise. OWASP's guidance on excessive agency is direct about who makes that kind of call: "utilise human-in-the-loop control to require a human to approve high-impact actions before they are taken." Re-sequencing a plan around a blocker that changes the finish date is exactly that kind of high-impact action — it belongs with the named lead, not with whichever pass produced the original sequence.

A worked example: a community theatre's spring production

A community theatre's production manager has a sliced task breakdown for a spring play: cast auditions, rehearsal blocks, set construction, costume fitting, lighting design, program printing, and a final dress rehearsal before opening night. She asks an AI assistant to draft a dependency map from the task list.

The assistant proposes sensible ordering dependencies: auditions before rehearsals, a final cast list before program printing, set construction and lighting design both before the dress rehearsal, dress rehearsal before opening night. It treats set construction and lighting design as independent — nothing about their task names suggests either depends on the other.

The manager checks the draft against what the AI assistant could not know. Costume fitting, she confirms, genuinely depends on two things the task names alone do not show: the final cast being confirmed, which the AI draft already caught, and fabric arriving from a supplier with a three-week lead time — a real external blocker that changes when fitting can actually start, and one only she has visibility into. She also corrects the "independent" call on set construction and lighting design: both jobs are done by the same one part-time volunteer, who cannot work on both at once, so in practice they share a dependency the AI assistant had no way to see from the task list alone. She reorders the plan around the fabric lead time and the shared volunteer before treating the sequence as settled.

Accessibility notes

This lesson is text-first, with no images, audio, video or downloadable artifacts. The practice exercise's model answer sits behind a native disclosure control that is reachable and operable by keyboard and correctly announced by screen readers. The knowledge check uses native radio-button inputs with a visible question and options, and posts its result to a live status region so assistive technology announces the outcome without a page reload.

Sources and limits

This lesson synthesises the sources below into a practical learning model. It is not a security standard, legal advice or a guarantee that any particular agent design is safe.

1. Project management glossary — APM (Association for Project Management). Defines a dependency as a relationship between activities in a network diagram, and a network diagram as a model of activities and their dependencies used in scheduling.
2. Project management glossary — APM (Association for Project Management). Defines a critical path as a sequence of activities through a precedence network from start to finish, the sum of whose durations determines the overall duration.
3. AI Risk Management Framework 1.0 — NIST AI Resource Center. Frames an AI system as an engineered system that generates outputs such as predictions, recommendations or decisions — not a verifier of which proposed dependencies are actually real.
4. LLM09:2025 Misinformation — OWASP Gen AI Security Project. Defines overreliance as placing excessive trust in AI-generated content without verifying its accuracy.
5. LLM06:2025 Excessive Agency — OWASP Gen AI Security Project. Recommends human-in-the-loop control requiring a person to approve high-impact actions before an LLM-connected system takes them.