A governed framework for AI-augmented project management the-mirr.com
“Most project managers work from memory. The MIRR AI Project Management framework lets them work from evidence instead, act as a subject matter expert on every project they run, and carry at least three times as many of them at once.”
initializemaprefreshreport

Run the first two once, when a project starts. Run the last two for as long as the project lasts.

What this is

A way of running projects in which an AI framework is wired into the tools your team already uses: the ticket system, the group chats, the email, the calendar, the document folders. It does the administrative half of project management, the half that eats the week, and it asks your permission before it changes anything, anywhere.

It is four commands. You run one, it proposes what it wants to do, you approve or you do not. Nothing happens on its own.

Who it is for

Project and program managers who already run several projects well, and the teams they sit in. Nothing here replaces what a good manager does.

It is for the point where the method runs out of hours. Every manager has a number of projects they can carry before depth starts costing breadth, and something quietly slips. MIRR moves that number, and holds the same standard across all of them.

The problem

The reconstruction tax

There are two costs in a project manager's week, and only one of them gets talked about. Administrative overhead is the obvious one. Ticket hygiene, meeting minutes, status updates, chasing owners for dates. Every project manager sees that one coming and budgets for it. It takes roughly 40 percent of the week.

Reconstruction is the expensive one, and nobody budgets for it at all. Another 30 percent goes to rebuilding the picture each time you switch projects. The information exists. It sits across a ticketing system, three chat channels, a document library, a mailbox and a calendar, and none of those systems knows the others exist.

Overhead and switching crowd out strategy and relationships, which was the only part that needed a human. That is not a discipline problem. It is an architecture problem.

The cycle initialize once · map once · refresh daily · report weekly
01 initialize ONE COMMAND

One shape, every program

Starts a project from nothing. You hand it the scope document and it builds everything the project needs before any work begins.

project folder skeletondocument repositorymeeting-minutes notebookproject chat channelthe ticketing projectthe first ticketskick-off message to the team

Each project gets its own isolated context, so two projects never bleed into one another.

02 map READ ONLY

Where the truth lives

Tells it where this project's information actually lives, and writes that down as the project's source list.

the ticket boardthe project chatsthe calendarmailbox threadsdocument folderswho the people are

You confirm every source before it is trusted, because a wrong channel means the wrong people get updated later. This step only reads. It changes nothing.

03 refresh GATED

The daily workhorse

The one you run every day. It visits every mapped source, reads everything new since last time, works out what actually changed, and comes back with a list of things it wants to do.

update this ticketsend this recapemail the attendeesflag this new riskclose this finished epicthe scope moved, is that real?

Each one shows its reasoning and where it came from. Then it stops and waits for you.

04 report WEEKLY

Status from evidence

Turns all of that into the status your leadership actually reads: how the project is doing, what got done, what is at risk, what is next, and what needs a decision from them.

Same shape every week, for every project, built from what happened rather than from what you remember on a Friday. A line that would not make an executive act does not survive the draft.

Initialize and map happen once per program. Refresh and report run for as long as the program does, and the cycle re-enters at map whenever the sources change.

What changes for the person, and for the function

For the project manager

The week invertsThe hours that went to ticket hygiene, minutes and status become minutes of reviewing and approving.
Follow-ups stop slippingRecaps, action items and owner nudges go out the same day as the meeting, because drafting is no longer the slow part.
Friday stops being an archaeology exerciseStatus is assembled from what the systems actually recorded, so you are editing a draft rather than reconstructing a week.
You stop being the lookup serviceAnyone asking where a project stands gets a current answer that did not require you to remember it.
The ceiling movesHow far depends on the portfolio. In the practice it was built in, it went from five concurrent programs to sixteen at the same hours.

For the PMO

Projects become comparableEvery project has the same structure and the same status shape, so a portfolio view is a real view rather than a collage.
Reporting arrives without chasingLeadership gets every project on the same day in the same format, and nobody spends Thursday collecting it.
Knowledge stops leaving with peopleThe project's context is written down as work happens. When someone moves on, the handover is the project, not a meeting.
New PMs start currentThey inherit a structured, up-to-date project instead of a folder, a rumour and three months of catching up.
Governance is uniform, not personalThe same approval and audit behaviour applies to every project, rather than depending on how disciplined each manager happens to be.
And what it costs Approving is real work, and there is a lot of it before the rhythm settles. Mapping the sources takes genuine thought, and a lazy map produces confident nonsense for weeks. None of this removes the need for a good project manager. It removes the reasons a good one runs out of hours.
The gate

Nothing reaches a real system without an explicit yes

An approval gate is a stop. The framework shows you the exact thing it intends to do, in full, and does nothing at all until you say yes. Not a setting that was switched on and could be switched off, but the way the system is built. A single daily run stops six to eight times. It is repetitive on purpose, and the repetition is what makes the promise real rather than a claim.

GATE 3 / 7 AWAITING APPROVAL
Post meeting recap to the project chat
sourcetranscript 09-18, 00:41:12 targetzero-trust-rollout (confirmed at mapping) reason3 decisions and 2 owners recorded, not yet circulated

It is asking, not acting. Approved is not committed until you press go. State transitions are gated the same way, not only messages.

GUARANTEE 01 Approval gates

Every write, message, ticket and status change. You see the exact payload, then approve, modify or reject.

GUARANTEE 02 Read-back verification

After every write it reads the record back and compares. What you approved is what was written, or you are told it was not.

GUARANTEE 03 Citation discipline

Every factual claim resolves to a source: a ticket and field, a transcript and timestamp. A claim that cannot be traced does not ship.

Governance the questions a security team asks first
Managed context store

Project knowledge lives in an environment the company controls. Not in an uncontrolled external repository, and not on the operator's own computer.

Swappable model layer

Which model runs it is a separate decision, tuned to your risk posture: cloud under contract, in-region private deployment, or fully self-hosted for air-gapped work.

Gates on every write

No outbound action, message, email or ticket change without explicit human authorization. A design constraint, not a setting to relax when someone is busy.

Complete audit trail

Every action proposed and every action executed is recorded, and the sequence is replayable for a compliance review or an incident.

Least privilege by inheritance

The framework reaches only what you are already authorised to reach. It inherits your access. It never expands it.

Adaptable to your frameworks

ISO 27001, SOC 2, TISAX, BSI expectations. Tuned to the posture you already run, rather than imposed on top of it.

Limits

What it does not do

Anyone offering you AI with no limits is offering you something else. An honest account of the boundaries is what makes the capabilities believable.

Failure modes, and where each is caught
A wrong source caught at mapping confirmation
A bad proposal caught at the approval gate
A claim with no evidence caught by the citation rule and the read-back
It does not decide It prepares decisions. A human makes them.
It owns no relationship Meetings, trust, hard conversations, negotiation. Human, always.
It never acts unsupervised No write without approval. There is no autonomous mode to switch on.
It is not a system of record It reads from and writes to the tools you already have. The truth stays where it always was.
It inherits your sources Bad in, bad out. Which is exactly why mapping has a human confirmation step.
Will it work where I am? three things that decide whether it survives a real company
It is not tied to any one product

Every step says what needs to happen, such as “add a comment to the ticket,” and never which product does it. Setting it up at a company means pointing each step at whatever that company actually runs. Jira or something else, Teams or Slack, it does not care.

It tells you up front what it cannot do

Before a command runs, it checks whether your tools can support every step of it. Anything missing is reported before you start, never discovered halfway through. Automation that stops in the middle of a live system is worse than automation that refused to begin.

Set up once for the company, then per person

The company setup happens once: which tools you use, how the framework reaches them, and what your organisation calls the levels of its work. Every manager after that is a short personal setup on top of it, not a rebuild.

If you want to try it one team · a handful of programs · one quarter

A 90-day shape that proves it, or does not

No rip-and-replace, no new system of record, no change to the tools you already run. The shape of the pilot is set out here so you can judge it, and argue with it, before you talk to anyone. Including me.

WEEKS 1 · 2 - STAND UP

Stand up the context architecture and map the sources. Agree the governance and data-classification posture with your security team before anything runs.

WEEKS 3 · 8 - RUN AND MEASURE

Run the daily refresh and the weekly reporting on the pilot programs. Measure against the baseline you took in week one. Tune in flight.

WEEKS 9 · 12 - DECIDE

Review the evidence and decide: scale, adjust, or stop. A clean decision either way, made on measurements rather than on a feeling.

Agree these five up front
hours reclaimed per PMreporting cycle timereporting consistencystakeholder-response latencysecurity sign-off
Exit criteria Worst case, one quarter and one team spent. You keep the structure and the learnings either way.