How I think.
A short companion to the dossier. The five phases below are not a framework I read in a book. They are the loop I have actually run on site, against the same constraints the dossier names. One walk, one statement, one markup, one walk-back, one SOP. Every artifact is priced in labor saved.
Observe.
Walk the site. Watch the work. Name what is missing. No deck, no model, no second-hand summary.
The first hour on a new job is the most valuable hour. I do not bring a checklist. I bring a phone, a sketchbook, and an unfilled head. I want the site to surprise me, because the things that surprise me are the things the drawings did not capture.
On a recent kitchen tenant improvement, three of the surprises became the three risks for the whole job. A type I hood with no coordinated makeup air, a walk-in subfloor that would not slope, and a grease interceptor route that crossed a column footing. Each one was a future change order. None of them were on the drawings I had been given.
Observation is not passive. It is the most active phase of the whole loop. The artifact at the end is a marked-up site plan with numbered pins, three to five of them, no more. If I cannot get it down to five, I have not understood the site.
Define.
Compress every observation into a single sentence. If you cannot, you have not observed enough. Time-box: one paragraph, ten minutes.
Defining is the cheapest phase and the most skipped one. The temptation is to keep collecting. The discipline is to stop and write one sentence.
The sentence has three parts. The thing that is broken. The constraint that makes it broken. The window in which it must be resolved. If any of the three is missing, the sentence is not yet finished.
On the kitchen job, the sentence was the column. Three trades, three routings, no shared drawing, eleven days. Once that sentence existed, everyone could carry it. The plumber could say it back to me at lunch. That is the test.
Prototype.
Build the cheapest version that can be touched. A sketch on a printed drawing. A one-page mock. A markup, not a model.
Prototypes in the field do not need to be CAD. They need to be unambiguous. I draw the new route over the existing route in orange marker on a printed sheet. Two callouts, two numbers. Lift mech four inches over the column. Drop plumb three inches under.
The prototype is wrong on purpose. It is a starting point, not a deliverable. The point is to get something in front of the trade who has to build it, so they can correct me. A markup the foreman can hold and circle is worth more than a model nobody opens.
If a prototype takes more than an afternoon, it is too expensive. If it takes more than an hour, it is probably too detailed for the question it is answering.
Test.
Hand it to the actual user. Watch what they do, not what they say. The senior trade is the user; the foreman is the user; the field super is the user.
Testing in the field is a walk. I walk the prototype with the trade who has to build it. I am quiet. They talk. They circle things, cross things out, ask the question that was not asked yet. Their hand on the page is the data.
The mech foreman approved the lift but flagged the hanger spacing at the splice. The plumber wanted a clean-out tee at the column. Electrical did not need to change anything. All three of those notes came out of a fifteen-minute walk. None of them would have surfaced in an email.
The output of the walk is a clean log with three states: approved, change, approved-with-note. No long prose. Three lines is enough. The next prototype is built directly from those three lines.
Ship.
Standardize the version that worked. Write the SOP. Train the next crew on the artifact, not on a verbal explanation.
- 01Mark column locations on the routing set before any trade rough-in.
- 02Mech sets first. 4-inch sleeves at every column line. Hanger spacing 4 ft on center.
- 03Plumb second. 3-inch drop with a clean-out tee at each column.
- 04Electrical last. Conduit stays at slab.
- 05Foreman signs each column before pour. No exceptions.
Most field knowledge dies at the end of the job. The foreman who solved the column went home. The next foreman, on the next job, gets the same problem and has to solve it again from scratch. That is the tacit knowledge tax the productivity research keeps naming.
The fifth phase pays the tax down. I write the resolution as a one-page SOP with five numbered steps. I attach the marked-up prototype as the picture. I hand it to the next crew before they hit the site. The training takes ten minutes because the artifact does most of the work.
Six crews have used the column SOP. Zero rebuilds since. That is the only metric that matters: the labor never spent again on a problem already solved.
Walk, write, sketch, walk back, ship the SOP. The loop is the product.
This is what an actually-useful field copilot would do, and this is the loop the NavigateAI pillars map directly onto. Quality control is observe and test, automated. Upskilling is the SOP, served at the moment of need. Scoping is the prototype, in minutes instead of days. Knowledge on Demand is the SOP library, kept alive across the portfolio.