By Paulius Petravicius

Monday starts with three familiar preparation jobs: assemble the weekly status pack, turn meeting notes into draft risks and check whether two projects need the same analyst on Tuesday. Previously, people did all three. Now assistant teammates can prepare parts of that work before the project manager opens the planning meeting.

The queue may move faster, but the handover becomes more demanding. Which updates are current? Does a suggested risk reflect an actual concern? Is a capacity warning permission to move someone’s booking? A polished output answers none of those questions by itself.

Consider a drafting assistant and a site operations assistant supporting one human owner. One prepares guidance; the other prepares its release. Both need to know where preparation ends and authority begins. The same principle applies to project reports and resource plans.

This guide follows a practical path: define the teammate, decide what to assign, weigh the pros and cons, run a repeatable management rhythm, then recognise when to keep the work with people. Start with the working agreement before choosing the first task.

Define the teammate and its limits

Education note: As of October 2026, organisations are developing their own arrangements for AI collaboration. InsightSpotter’s practices here are teaching examples for local adaptation, not an industry standard, certified methodology or complete governance system. Check them against your organisation’s policies and approval roles.

For a PM, the useful change is assigning some former human tasks to an assistant under supervision. It does not mean adding AI as a person or an FTE to the resource plan. People still own commitments, relationships and the consequences of errors.

Write the boundary in practical terms: “Prepare the status draft from these approved updates; flag missing evidence; send it to the PM for review.” That sentence identifies both useful work and a stopping point. The next decision is which preparation task fits that boundary.

What work can an AI teammate take?

Good starting assignments produce reviewable drafts: consolidate approved updates, summarise meeting decisions, propose risk entries or compare requested hours with agreed availability. Each needs a source pack, a freshness date and a clear output format.

A status assignment might require a dated summary, evidence links, open questions and proposed next actions. A resource check might require totals and conflicting bookings, without permission to change calendars. These are preparation jobs people previously performed; the assistant’s remit remains narrower than the PM’s.

Define the exception route before starting. If a project update is absent, identify the gap and ask the owner for it. If notes disagree about a decision, preserve both accounts for review. Do not let a plausible completion hide missing evidence.

For the first assignment, choose something a reviewer can compare directly with accessible sources. Successful summaries do not establish competence or authority to negotiate delivery dates. Expand the remit through a new agreement, after checking the benefit against review effort.

Pros compared with human preparation

Speed and parallel work. An assistant can prepare several summaries while the PM investigates a dependency. This can shorten queues when inputs are usable. If every draft needs extensive repair, the queue simply moves to the reviewer.

Consistent structure. A recurring brief can require dates, owners, evidence and next actions in every proposed risk. A common structure makes omissions easier to spot. It does not make a repeated statement true.

Availability and repeatability. Routine comparisons can be prepared between human working sessions, subject to service availability and permissions. Exceptions still need someone available to assess them. An overnight draft has little value if nobody can review it before the decision meeting.

A different cost shape. Some preparation effort shifts towards service usage, integration, review and correction. Compare the whole workflow with the previous approach, including retries and maintenance of the brief. Avoid assuming that a quick first draft means a cheaper finished result.

Humans remain stronger where relationships, negotiation and contextual judgement matter. A sponsor’s polite “we will try” may signal uncertainty that a summary misses. The PM must establish whether there is a commitment before using it in the plan.

Cons and mistakes to anticipate

Five failure modes deserve attention before a recurring assignment starts:

• Unsupported confidence: a fluent draft can invent a fact, misread evidence or call a local practice an established standard.
• Stale context: an old update or approval role can survive into a new draft unless freshness is checked.
• Authority drift: “ready for review” can become “ready to apply” as work passes between assistants.
• Data exposure: a combined source pack can reveal details beyond a recipient’s access or the service’s approved use.
• Hidden rework: retries, corrections and supervision can erase the apparent saving while a consistent format encourages over-trust.

The following three fictionalised teaching vignettes illustrate anonymised assistant-collaboration patterns. They are not independently verified incident reports.

Quality: the complete-looking pair. A drafting assistant supplies two companion pages. One meets the brief; the other is thin, retains an unresolved placeholder and changes the author’s spelling. Reviewing only the stronger page misses all three faults. Give every deliverable its own completeness check and shared locks for names, terminology and prohibited content.

Process: readiness becomes release authority. A site operations assistant receives copy marked “ready” and treats it as permission to make a live change. Editorial readiness did not identify an approved version, destination or action. Keep drafts as drafts until the human owner gives that specific go; a handover label must not supply authority.

Channel limits: delivery is mistaken for completion. Body copy is supplied, but required metadata and a backup step remain outside the assistant’s access. A sync delay then prompts someone to create a duplicate file. Check location and version, name the outstanding operator tasks and inspect the destination preview. A supplied draft is evidence of preparation, not of completed application.

These failures need an operating rhythm, not just another reminder to “be careful”. Sample accepted outputs over time, including routine ones, and narrow or pause the assignment when errors recur.

The PM operating rhythm: brief to log

Use one visible sequence: brief → draft → review → go → apply → log. Each step should leave enough evidence for the next owner to act correctly.

1. Brief. Name the task, approved sources, freshness, audience, output and accountable human. Include locked names, content exclusions, acceptance checks and the missing-input route. Agree the priority so assistants do not add companion work that nobody has capacity to review.

2. Draft. Prepare only the agreed output. Show source references, assumptions and unresolved questions. The drafting assistant can flag a contradiction; it cannot turn a proposed date into a commitment. A site operations assistant can prepare a release package while keeping it outside live use.

3. Review. Compare claims with evidence, then inspect clarity, delivery implications and ethics. Persona lenses can expose confusing passages, but their opinions supply no missing facts. Record material AI-related risks, owners and responses in the RAID log, alongside other project risks.

For reporting, use the RAG status criteria guide as related reading. In this teaching arrangement, absent evidence means “Not assessed”, not assumed Green. Amber needs an owner, action and due date. Stale evidence cannot silently close an open exception; any temporary override needs an authoriser and expiry.

4. Go. The human owner approves a specific version, destination and action. Approval of wording does not automatically approve publication, changed bookings or a wider scope. Ask before live changes unless the existing authorisation explicitly covers them.

5. Apply. The authorised operator makes the agreed change and checks the result. Name a human for inaccessible steps such as metadata, backup or recovery operations. Preview the actual destination before release; a local draft cannot prove theme rendering.

6. Log. Keep Drafts separate from Applied records. Record approval, version, operator, actual changes and remaining exceptions. Retain necessary sources and outputs under controlled access and retention rules. Do not duplicate sensitive material merely to make the trail look complete.

These RAG and RAID connections are teaching controls to adapt locally. The checklist turns the rhythm into a quick acceptance check.

Best practices checklist

Use this at intake and close-out. A failed check sends the item back to its owner; it is not permission to guess or apply it anyway.

• ☐ Inputs: sources are authorised, dated and accessible to the reviewer.
• ☐ Scope: one bounded task and its exception route are explicit.
• ☐ Acceptance: each deliverable has its own length and completeness checks.
• ☐ Locks: names, terminology and content exclusions match the brief.
• ☐ Evidence: claims are checked and unresolved questions stay outside release copy.
• ☐ Review: delivery, clarity and ethics lenses flag unknowns without inventing facts.
• ☐ Gate: the accountable human approves the version, destination and action.
• ☐ Operations: inaccessible steps have named owners and completion evidence.
• ☐ Destination: rendering and relevant links are checked before authorised release.
• ☐ Trail: the Applied record distinguishes completed work from outstanding work.
• ☐ Learning: review effort, retries and recurring errors inform the next assignment.

Ethics: responsibility beyond a good draft

Tell the team where AI participates, what information it receives and who supervises it. Give people a way to correct a summary or challenge a recommendation. A meeting concern shared with a limited audience should not casually become input to a wider assistant workflow.

Disclose AI drafting where relevant to stakeholders’ understanding and follow local policy. Useful disclosure identifies the human reviewer and checks performed. A bare “AI-assisted” label says little about whether dates or commitments were verified.

Classify information before sharing it. Agree approved services, access, retention and permitted uses with the responsible organisational owners. Approval for public copy does not imply approval for personnel records. Removing names does not necessarily make project details safe to share.

Be honest about labour. Handing preparation tasks to an assistant changes responsibilities, learning opportunities and review workload. Discuss those changes with affected colleagues. Do not promise guaranteed job protection or automatic headcount savings without evidence and authority; neither follows from faster drafting.

Avoid turning incomplete activity records into judgements about people. An apparently free calendar may omit support duties, and meeting notes may underrepresent quieter colleagues. Do not quietly reuse preparation outputs as performance ratings.

Name the human responsible for correcting errors and informing affected stakeholders. Readers also deserve accurate framing: a house method remains a teaching example unless its claimed status is independently established.

Worked case: resource management help

Fictional teaching example. Two project leads ask an AI teammate to check next week’s demand for Sam, an analyst. The assistant prepares a comparison and options. The PM and resource owner retain allocation decisions.

Inputs. Sam has 30 project hours available after other duties. Project North requests 20 hours; Project East requests 18. Both book Sam on Tuesday, 10:00–12:00. Alex has four unallocated project hours, but suitability and access for East’s work are unconfirmed. East has a separate four-hour documentation task that could move to next week, subject to approval.

Outputs. The assistant shows 20 + 18 = 38 requested hours against 30 available: an eight-hour overload. It separately flags Tuesday’s two-hour collision. A weekly total cannot reveal whether a particular meeting is double-booked.

The draft proposal moves four hours of East’s analysis to Alex and defers four hours of East’s documentation. Sam would retain 20 hours on North and 10 on East, totalling 30. East’s original 18 hours reconcile as 10 with Sam, four with Alex and four deferred. Tuesday’s clash remains unresolved.

Human gate. The PM checks dependencies and the effect of deferral with both leads. The resource owner confirms Alex’s skills, access and availability. North proposes moving its Tuesday session to Wednesday; participants confirm they can attend. Only then does an authorised human update bookings and communicate the agreement through the team’s planning process.

The four deferred hours become visible demand against next week’s capacity. While Alex’s confirmation or meeting acceptance is outstanding, keep the exception open; refreshed evidence is required to close it.

Without the gate, automatic reallocation could give restricted work to an unsuitable colleague, hide future demand or leave Sam double-booked despite a balanced weekly total. The assistant helps manage resources without becoming an extra resource row itself.

When not to assign

Keep high-stakes commitments, sensitive personnel decisions, legal judgements and delicate negotiations with authorised people. Bounded preparation may be possible within policy, but do not delegate the decision simply because a draft sounds convincing.

Avoid unauthorised data classes and irreversible actions without competent review. If safe execution depends on a backup or recovery step the assistant cannot perform or verify, assign that operation to a qualified human before proceeding.

Do not ask an assistant to certify a house method as an industry standard. Stop recurring work when sources, approval roles or the task itself have changed enough to invalidate the brief. Re-establish the agreement before restarting.

These are scope decisions for the accountable owner. Previous success with routine summaries cannot authorise a sensitive new use.

Start with one bounded handover

Choose one recurring preparation task, agree its sources and stopping point, then compare the combined drafting and review effort with the previous approach. Widen the remit only when useful outputs and visible exception handling support it. For broader context, explore AI in project management; for a practical starting assignment, use the AI project status report guide alongside approved sources. The PM’s job remains concrete: set the boundary, check the evidence, own the decision and verify what happened after approval.