The project is forty-one days behind, the owner wants a recovery plan by Friday, and the project manager points the AI scheduling tool at the problem on Wednesday morning. By the afternoon it has drafted three recovery options: a re-sequencing that recovers eighteen days by reordering the interior fit-out, a crashing option that recovers twenty-six days by doubling the drywall and MEP crews at a premium, and a fast-tracking option that recovers thirty-one days by overlapping the remaining structure with the early skin, each with its cost and risk laid out. The plan is drawn, it is internally consistent, and it would be easy to print it, hand it to the owner, and call it the recovery. But a recovery schedule is not a drawing; it is a set of commitments, and a beautifully drawn recovery that doubles the drywall crew is worth nothing if the drywall sub cannot actually field a second crew, and a fast-track that overlaps structure and skin is a liability if the trades have not committed to the overlap and the safety plan for it. This lesson is about using AI to draft recovery options fast while keeping the selection, the verification, and above all the real trade commitments behind the recovery firmly with the human, because a recovery schedule that re-directs the entire project is among the most consequential documents a PM publishes, and one drawn without commitments behind it is a fiction that fails twice, once when it is published and again when the work does not recover.

AI Drafts the Recovery Options, the Human Decides the Recovery

When a project falls behind, recovery means finding a way to claw back the lost time, and the classic levers are well known: re-sequencing the remaining work into a more efficient order, crashing activities by adding resources to shorten their duration, and fast-tracking by overlapping activities that were planned in sequence. Each lever recovers time at a cost, crashing buys speed with money and congestion, fast-tracking buys speed with risk and rework exposure, re-sequencing buys speed with the disruption of changing the plan, and the recovery is some combination of the three. AI is well suited to drafting these options, because, like ALICE generating build scenarios, it can explore the combinations of levers across the remaining schedule and return several recovery options with their trade-offs computed, far faster than a planner working the logic by hand.

This is the generative engine in its drafting role, the same role it plays drafting a submittal log or an RFI response or a build scenario: it produces candidate options fast, surfacing the trade space of recovery so the PM can see what eighteen days of recovery costs versus twenty-six versus thirty-one. The value is genuine and it is the value of speed and exploration: a PM who would have spent two days hand-building one recovery option gets three options in an afternoon, with the costs and risks laid out, which is more options and more analysis than the manual process could produce against the Friday deadline. The decide-then-draft principle runs exactly as it did with ALICE: the options are candidates, the drafting is cheap, and the human's selection of which recovery to commit to is the deliberate act that carries the responsibility.

So AI drafts the recovery options and the human decides the recovery, which is the same division the program has taught across the generative engine's applications. The options are not the recovery; they are the menu from which the PM selects the recovery, and the selection is consequential because the chosen recovery will re-direct the project, re-loading the crews, re-ordering the work, and re-setting the commitments. The PM owns that selection, and the AI's role ends at surfacing the options the PM chooses among, which is the drafting role, valuable for the speed and the exploration but stopping short of the decision the human must own.

A Recovery Schedule Re-Directs the Project, So Verify the Logic and Feasibility

A recovery schedule is one of the most consequential documents a project produces, because it does not just describe the work, it re-directs it: it changes the sequence, re-loads the crews, commits the premium money for crashing, and accepts the risk of fast-tracking, and once published it drives the next phase of the project as forcefully as the original baseline did. So the recovery schedule sits squarely at the schedule gate, the same gate the original baseline cleared, and the cardinal rule applies: verify before the recovery drives the work. A flawed recovery does not just fail to recover; it re-directs the project into a worse position than the delay it was meant to fix, spending the premium money and accepting the fast-track risk for a recovery that does not materialize, so the verification is proportioned to this high consequence.

Verification of a recovery option has two parts: the logic and the feasibility. The logic check asks whether the recovery is internally sound, whether the re-sequencing respects the real dependencies (you cannot fast-track the finishes ahead of the rough-in they cover), whether the crashed durations are achievable (doubling the crew may not halve the duration if the work face cannot hold two crews), whether the fast-track overlaps are physically possible (overlapping structure and skin requires the structure to be far enough along to hang the skin). The AI's drafted option can be logically flawed in exactly the way an ALICE scenario can, optimizing the recovery within the model while violating a dependency the model did not hold, so the logic check hunts for the dependency the recovery quietly broke to recover the days.

A recovery schedule is a promise the project makes to itself and the owner that it can claw the time back, and a promise drawn without the commitments to keep it is worse than no promise, because it spends the premium and the risk on a recovery that was never real and delays the honest reckoning until the recovery fails.

The feasibility check asks whether the recovery is buildable in the real world, which is where the trade commitments enter, and it is the part the AI most cannot supply. The AI can draft a recovery that doubles the drywall crew, but whether the drywall sub can actually field a second crew, whether the labor is available in the market, whether the GC can stage two crews on the work face, are feasibility questions the model does not hold and the PM must verify with the subs before committing the recovery. This is the consequence-proportioned verification the program teaches, applied to the highest-consequence schedule document: the recovery re-directs the project, so it earns the deepest verification, both the logic check that the recovery is sound and the feasibility check that it is buildable, before it drives the work.

The Trade Commitments Behind the Recovery Must Be Real, Not AI-Drawn

Here is the heart of the lesson, the point that separates a real recovery from a paper one: the recovery is a set of commitments, and the commitments must be real, made by the trades who will deliver them, not merely drawn by the AI. When the recovery option says the drywall crew doubles in week six, that is not a fact the AI can assert; it is a commitment the drywall sub must make, and until the sub has agreed to field the second crew, the recovery's eighteen recovered days are a number on a page with nothing behind it. The AI can draw the doubled crew into the schedule, but it cannot make the sub commit to it, and a recovery whose commitments exist only in the drawing is a fiction that will fail when the second crew does not appear.

This is why the recovery must link to the commitment process, which the next lesson develops as Last Planner pull planning: the recovery's levers, the crashed crews, the re-sequenced work, the fast-track overlaps, must be converted into commitments the trades actually make, in a planning session where the subs agree to the crew loading, the sequence, and the overlap, with their constraints surfaced and resolved. The AI-drafted recovery is the proposal that goes into that session, and the session is where the proposal becomes a recovery, because the trades commit to the levers and the recovery becomes real. Without the session and the commitments, the AI-drafted recovery is a unilateral plan handed to trades who never agreed to it, and people do not deliver on plans handed to them, they deliver on commitments they made, so the recovery handed down fails where the recovery committed to holds.

The distinction is the difference between a schedule and a recovery. A schedule can be drawn; a recovery must be committed, because a recovery asks the trades to do something harder than the original plan, to surge crews, to work out of the comfortable sequence, to accept the fast-track risk, and they will only do the harder thing if they have committed to it, not if it was drawn for them. So the PM's central job in the recovery is not to perfect the AI's drawing but to convert the drawing into real commitments, taking the AI-drafted options into the trades and securing the agreements that make the chosen recovery something the project can actually deliver, which is the work the AI cannot do and the work that determines whether the recovery recovers.

The Human Owns the Recovery Plan and the Commitments

The PM holds responsible charge of the recovery plan, the same way the scheduler holds responsible charge of the baseline, and for the recovery the charge is heavier because the recovery is more consequential and more dependent on commitments the human must secure. The AI drafts the options, but the PM owns the selected recovery: they answer to the owner for it, they are accountable for the premium money the crashing spends, they carry the risk the fast-tracking accepts, and they are responsible for whether the commitments behind it are real. The AI cannot be in the owner's meeting defending the recovery, cannot be accountable for the premium, cannot secure the trade commitments, so the PM holds all of it, and the AI's drafting does not transfer any of the ownership.

Owning the recovery means owning both the plan and the commitments, and the two are inseparable, because the plan is only as real as the commitments behind it. A PM who owns a beautifully verified recovery plan but has not secured the trade commitments owns a fiction; a PM who has secured the commitments owns a recovery. So the ownership extends past the plan to the commitments, and the PM's accountability is for both, which is why the PM cannot delegate the recovery to the AI's drawing, because the AI can produce the plan but not the commitments, and the commitments are where the recovery lives. The responsible charge is for the recovery as a committed reality, not the recovery as a drawn schedule.

This ownership also means owning the honesty of the recovery. A recovery drawn to satisfy the owner's Friday deadline but without real commitments behind it is a dishonest recovery, a plan that claims to recover the time while knowing the commitments are not there, and the PM who publishes it owns the dishonesty and the failure that follows. The honest alternative, when the commitments cannot be secured, is to tell the owner that the time cannot be fully recovered with the available resources, which is a harder conversation than handing over an AI-drawn recovery but the one the responsible charge demands, because owning the recovery means owning the truth of whether it can be delivered, not just the drawing that claims it can. The PM owns the recovery plan and the commitments, and owning them means making the recovery real and honest, not just drawn.

The Applied Problem: A Recovery-Planning Protocol

Here is the exercise. Build the recovery-planning protocol your project will use when it falls behind and needs a recovery schedule, with the AI drafting the options and the human owning the selection, the verification, and the commitments. The protocol's purpose is to convert a delay into a real, committed recovery, using the AI's speed to generate and analyze the recovery options while keeping the consequential selection, the logic and feasibility verification, and the real trade commitments with the human, so the recovery the project publishes is one it can actually deliver.

Produce two artifacts. First, the recovery-planning protocol: the AI's drafting role (generating the re-sequencing, crashing, and fast-tracking options with their cost and risk trade-offs), the PM's selection of which recovery to pursue, the two-part verification the selected recovery clears (the logic check that the recovery respects the real dependencies and achievable durations, and the feasibility check that it is buildable with the available resources), and the commitment step that converts the selected recovery into real trade commitments before it is published as the recovery. Second, the commitment analysis: why the recovery's levers, the crashed crews, the re-sequenced work, the fast-track overlaps, must be converted into commitments the trades actually make rather than left as AI-drawn numbers, how the recovery links to the Last Planner pull-planning session where the commitments are secured, and the rule that a recovery without real commitments behind it is not published as a recovery, because a drawn recovery without commitments is a fiction that fails twice.

The deliverable is the recovery-planning protocol and the commitment analysis, and the lasting product is a project that uses AI to draft recovery options fast while keeping the recovery real, the PM selecting the recovery, verifying its logic and feasibility, and securing the trade commitments that make it deliverable, so the recovery the project commits to is one the trades have agreed to and can actually achieve. The PM who masters this gets the AI's speed in generating and analyzing the recovery options, which matters against a Friday deadline, without the failure of a drawn recovery without commitments, because the protocol places the selection, the verification, and the commitments with the human, where the recovery is made real, and reserves for the AI only the drafting, where its speed is the value and its inability to commit is not yet a liability.

Crashing Versus Fast-Tracking: The Trade-Off the PM Owns, Not the AI

The three recovery options in front of the PM are not equivalent ways to buy the same time, and the choice between them is a risk decision the AI cannot make. Crashing buys time by adding resources to activities on the critical path, the doubled drywall and MEP crews that recover twenty-six days at a premium. Its cost is money and diminishing returns, because a second crew rarely doubles the output of the first, and at some point the work face is simply too congested to absorb more bodies. Fast-tracking buys time differently, by overlapping activities that were planned in sequence, the remaining structure overlapping the early skin to recover thirty-one days. Its cost is not money but risk, because work begins before its predecessor is complete, and if the early work changes, the overlapped work built on it has to be redone.

The AI can draw both options and compute their nominal day savings and their direct costs, and that arithmetic is genuine value, but it cannot weigh the risk the way the decision requires. Fast-tracking that overlaps structure and skin carries a rework exposure that depends on how settled the structural design is, how much float the trades have, and how a coordination miss would cascade, judgments that live in the PM's knowledge of the project and the trades, not in the schedule logic. The AI will report thirty-one days recovered as cleanly as it reports twenty-six, with no sense that the thirty-one are riskier days, bought with an exposure that could erase them if the overlap goes wrong. Treating the larger number as the better option because it recovers more time is exactly the error the metric-as-signal discipline warns against, reading the recovered-days figure as a verdict rather than a signal that still needs the PM's risk judgment.

So the choice among re-sequencing, crashing, and fast-tracking is the PM's, made on the project's actual risk tolerance and the trades' actual capacity, not on which option the model scored highest. Often the answer is a blend, some re-sequencing where the logic allows it, a measured crash where the work face can hold the crews, a limited fast-track only where the predecessor work is settled enough to overlap safely. That blended judgment, proportioning each lever to where it can be delivered without buying back the time in rework or congestion, is the responsible-charge decision at the heart of the recovery, and it is the part the AI's speed accelerates the inputs to but never replaces.

Key Takeaways

• AI drafts the recovery options fast: when a project falls behind, the classic levers are re-sequencing, crashing (adding resources to shorten durations), and fast-tracking (overlapping sequential activities), and the AI can explore the combinations and return several recovery options with their cost and risk trade-offs computed, far faster than a planner working the logic by hand.
• This is the generative engine in its drafting role, the same as drafting a submittal log or a build scenario: the options are candidates, the drafting is cheap, and decide-then-draft runs exactly as with ALICE, the PM's selection of which recovery to commit to being the deliberate human act that carries the responsibility.
• A recovery schedule is among the most consequential documents a project produces because it re-directs the work, changing the sequence, re-loading the crews, spending the premium for crashing, and accepting the fast-track risk, so it sits at the schedule gate and earns the deepest verification before it drives the work.
• Verification has two parts: the logic check (does the recovery respect the real dependencies and achievable durations, or did it quietly break a dependency to recover the days) and the feasibility check (is the recovery buildable with the available resources, which is where the trade commitments enter and the part the AI most cannot supply).
• The heart of the lesson: the recovery is a set of commitments, and the commitments must be real, made by the trades who will deliver them, not merely AI-drawn. A doubled drywall crew in the drawing is worth nothing until the drywall sub agrees to field the second crew, so an AI-drawn recovery without commitments is a fiction.
• The recovery must link to the commitment process (Last Planner pull planning, the next lesson): the AI-drafted recovery is the proposal that goes into the session where the trades commit to the crew loading, the sequence, and the overlap, because people deliver on commitments they made, not on plans handed to them, so the recovery committed to holds where the recovery handed down fails.
• The PM holds responsible charge of the recovery plan and the commitments, heavier than the baseline charge because the recovery is more consequential and more dependent on commitments the human must secure: the AI drafts but cannot defend the recovery to the owner, be accountable for the premium, or secure the trade commitments, so the PM owns all of it.
• Owning the recovery means owning its honesty: a recovery drawn to satisfy a deadline without real commitments is dishonest, and the honest alternative, when the commitments cannot be secured, is to tell the owner the time cannot be fully recovered, the harder conversation the responsible charge demands, because owning the recovery means owning the truth of whether it can be delivered, not just the drawing that claims it can.