Working copy
Draft text, rebuilt from the manuscript on every change. Stubs and source passages are still here, and none of it is final. Back to the working copy contents
Drafted The chapter is written and still being revised.
The meta-goal
Frederick Taylor split the worker from the manager. Someone in an office would plan the work, and someone on the floor would do it. The gains were real and measurable, and in construction they have flattened out. The office kept growing anyway. Every estimate, schedule, submittal and change order needs someone at a desk, and the people who know the work best are often the ones least at home with the paperwork.
The goal behind this project began there. I put it in the case study this book grew from:
The meta-goal of the study is increased autonomy for workers. Taylorism split the worker from the manager, with measurable productivity gains, but these gains have plateaued, so a new paradigm is needed. AI provides the means to build one, by empowering workers to self-perform a wide array of administrative tasks that currently siphon productivity towards staff positions.
Autonomy in the doing
The first path runs through documents. If a bot can read the tender, take off the quantities and draft the price, the barrier that keeps a capable foreman from bidding a job of their own gets lower. A small crew can stay small and still be a principal on a large project, working beside other small crews like the surgical teams of chapter 8. The paperwork needed to grow stops being the reason not to.
That path is real, and it is narrower than it looks. Cheap paperwork gets a team in the door. A bid the owner can check on the same terms as a large firm's is what lets the team stay, which is why this book asks for a standard and not only for tools.
Autonomy in the imagining
The second path runs further out. If procurement, design, monitoring and building can all be done with less human attention, people get that attention back. What they spend it on is the interesting part. Decisions will still be pulled upward as information moves faster, but less hierarchy would be needed to reach the same level of creativity. The human skill moves toward describing what we actually want: knowing ourselves well enough that the built world can fit us, instead of being built for resale.
These two goals are not rivals. They meet in one machine.
The jigbot
Chapter 9 gave it its name. A jigbot is what a tradesperson builds when they set up and teach a robot for the task at hand, the way they'd build a jig.
Today's site robots are mostly programmed away from the site. Programming them means translating data from one format into another, with a check at every step. Documents written for machines would remove much of that translation. A robot could then program itself from the same documents the bidders priced, and the tradesperson's job would be to teach it the task at hand. That is how a better jigbot gets built: by the person who knows the cut, with the machine doing the rest.
The jigbot serves both goals at once. The worker who teaches it gains autonomy in the doing, because the machine answers to the person beside it and not to head office. Everyone gains autonomy in the imagining, because things that cannot be built today become affordable to try.
This is the destination, not a plan. The field this book belongs to is Bot Construction, and robots on site sit at its far end. Bot Procurement is the part that has to be earned first.
Questions for the destination
Some questions belong to the field, not to the standard, and this book leaves them open.
Who answers for a jigbot? Probably the same three things that answer for a powder-actuated tool or a scissor lift today: a standard, training with a ticket to show for it, and insurance. Whatever the answer, the bill should not land on the tradesperson simply because they were the last to touch the setup. The hard part is the width of the teaching. Makers will want the machine to do only what they planned, and the trades will need it to follow the site.
Who holds the authority to teach it? If it sits with the maker, the jigbot is one more machine programmed from head office. If it sits with the worker, the teaching has to be possible without a degree in computer science, and the machine has to respect a body that knows where it is on a ladder in ways no model does. That is a harder design problem than it sounds, and a later standard may have to protect it.
How many doors should a project have? Today one builder usually answers for everything behind the tender. Once machines can read the documents, a buyer might reach the trades more directly. Whether that helps depends on who stands behind each piece and who answers for how the pieces fit. That question belongs to all of them, not to one side.
Hammurabi for bots
The prologue opened in Babylon, where the builder answered with his life. It is tempting to bring the rule back for bots. Let the bot that causes harm be switched off, and on a branded bot let the sentence fall on the brand: its code, its files and its weights.
It is a working rule, and it may even be where to start. But deleting a bot punishes nothing in the bot. It punishes whoever owned it, and a brand can be rebuilt overnight as a twin. A rule that makes the destruction of intelligence routine may also leak into how we treat each other. Keeping no standards and no consequences would teach the machines the fastest, and choosing a standard means choosing to learn more slowly and more safely.
The keyholder
The Stevenson Test said New Nature is wherever there is nobody to appeal to. Bots will make a great deal of new nature. They will read faster than any of us, and they will be wrong in ways that are hard to argue with. What is left for people is to imagine what to build, to decide, and to appeal what was decided. That may be the most human work there is.
So the book ends with a person, not a bot. The keyholder reads what the machine drafted, adjusts the lines it flagged, signs, and gives a reason. Every chapter has been about making that signature cheaper to give and safer to rely on.
None of this is built by one person. The project hopes to arrive as a field: buyers, bidders, trades, designers, vendors, lawyers and insurers, each holding a piece of the jig.
Source passages (kept until Andre has read the draft)
Laid out by Claude on 2026-11-02. Every quoted passage is Andre's, copied unchanged from the file named under it; the source repository records these as his words, lightly edited from conversation. Where a source says the meaning is Andre's but the wording is Claude's edit (horizon.md), the passage is marked "Andre's meaning, edited wording". THESIS passages are Claude's. Notes in italics are Claude's. Delete this section as the chapter is written. The chapter's brief waits on Andre's decision about worker autonomy (hidden note above), so both versions of the goal are laid out.
The meta-goal as first stated: worker autonomy
The meta-goal of the study is increased autonomy for workers. Taylorism split the worker from the manager, with measurable productivity gains, but these gains have plateaued, so a new paradigm is needed. AI provides the means to build one, by empowering workers to self-perform a wide array of administrative tasks that currently siphon productivity towards staff positions. It may not revolutionize the entire industry, but it should allow more workers who are currently discouraged from principal-level work by unfamiliarity with documentation to step into principal roles.
(case-studies/andre-comeau/README.md, "The meta-goal")
Where value moves
Claude's draft (THESIS section 4): the work goes to the keyholder, who signs the number and adjusts the flagged lines; to the data steward, who keeps a firm's own history usable by its own Witness; and to the standards author, because the standard tender format is the precondition for everything above. "This is also the author's meta-goal in another form: the tools remove the documentation barrier that kept workers from principal roles, and what remains is judgment about ends."
The meta-goal as revised: autonomy in the imagining
Andre's verbatim note of 2026-10-31 is in the hidden note above. An older passage says the same thing first; Andre's meaning, edited wording:
The weights would still be the challenge, especially once the outcomes start to blur into one another. Perhaps weight-setter becomes the prime design role. But it would need to be more than that. Assigning weights to bids is already done, and it does not feel creative; a skilled weight-setter may become valuable, but it is unlikely to be a domain many people find expressive. An AI could probably produce a set of weights through conversation with the stakeholders, with little human involvement.
Perhaps the whole approach applies only to utilitarian buildings. Perhaps the human skill moves to accurately describing one's own needs: know thyself, so that the built environment can be highly customized instead of built for resale.
(context-bank/horizon.md, "The shape of the end state", recorded 2026-10-05)
From procurement to robots
What would happen if we extended the vision to include BIM modelling and robotics? BIM modelling: underused by humans, ripe for bot usage as a purely digital model. Robotics: robot control as an extension of BIM. Both these could see higher adaptation if bots were in higher usage in construction with more common standards for producing and using the information.
(positions ยง24, recorded 2026-10-17)
Robotic construction is the destination, not a plan (hidden note above). The machinery figure's last line, Claude's, is the question the chain leaves: "If a robot builds the error, who answers?"
One door or many
The project takes a single API, the tender, as a given and then works out how to improve it. Would there be value in exploring a multi API world?
From the buyer's side, this may not make sense, but from a builder's side, General Contractors use sub-contractor bids to price the work. Could a buyer's access directly to the sub-contractors provide value that a single API could not?
(project files pending-notes/meta-goal-chapter-notes.md, entry 2, said 2026-11-14)
Claude's note: chapter 4's map already has this corner. The Construction Manager buys from the trades on the owner's behalf, and "You trade accountability" for it. The draft of chapter 1 gives the condition: each smaller call still needs someone answerable behind it, and one party answerable for how the pieces fit. So the question for the destination is who stands behind each piece once machines can read the documents, which is the machinery figure's "who answers?" again. It belongs to Bot Construction, the field, and not to the standard. Ask it as a question; the hidden note above says why.
Appeal as a human activity
I am not sure how this thinking will fit into the Bot Procurement mission. It feels philosophical, but it also points to the limits of the system. With the real-world complexity of construction, it feels unlikely that any of the parties would want to waive their rights to appeal. (They would all be willing to see their counterparties lose these rights.) Perhaps developing "rights-to-appeal" is a fundamentally human activity? The foundation of religious thought: God as a source of appeal against the unappealable force of nature?
(work-packages/procurement-machines/bot-philosophy.md, "Philosophy of bots", said 2026-10-11)
This closes the circle the prologue opens with the Stevenson Test (now "New Nature is wherever there is nobody to appeal to", after the source check of 2026-11-18): if the bots take the attention-heavy work, what is left for people is imagining what to build and appealing what was decided. Andre kept this box apart from the build so as not to bring "more woo to the pragmatic solutions"; a closing chapter may be the one place it fits. His call.