Back to the episodes

TK: ten posts

Ten posts for Tanmaya, from his 18 years of building and his own words on The Punch List and in interviews. Tap a first line to read the post. Under each one: where it came from, and what we'd check with him.

1As a structural engineer, I was guilty of the Ctrl C, Ctrl V problem.

As a structural engineer, I was guilty of the Ctrl C, Ctrl V problem.

A detail from an old project looked close enough, so I copied it into the new drawings, reviewed it and moved on, instead of drawing it for the building in front of me.

Then I moved to the contractor side and became the guy complaining about those same kinds of details. "This detail is not for this project. This detail doesn't even match the project. What the hell? What is this guy doing?" I knew exactly what he was doing, because I had done it too.

It made sense at the desk, because copying a detail takes the engineer seconds. But finding out it doesn't fit takes an RFI, a redesign and a lot of other people's time, and that cost shows up on site, in someone else's budget.

People keep asking how construction productivity could fall while our technology got better. I think one of the core reasons is that we let tools drive our process. We copy and paste because computers let us, instead of thinking through the details as we draw them.

Now AI can copy and paste at a scale I never could. I hope to God it doesn't become one more contributor.

So this time around, I start with the process. Before I buy any tool, I want to know the steps we take to get the outcome we want. Only then do I ask whether the tool makes those steps faster.

From TK's Beiing Human interview (13 Jul) and The Punch List (11 Sep).

To check: His own Ctrl C, Ctrl V story. Is he happy to tell it?

2"What the hell did you do?"

"What the hell did you do?"

That was the whole RFI process on my first job, designing power plants as a structural engineer. The engineer didn't just design the structure. We gave the material quantities, and I had to order the rebar myself.

If it didn't fit, there was no RFI. There was just that question, and it was coming to me.

I was really responsible for the material being right. When you order the steel yourself, you check the drawing like your name is on the truck.

Then I moved into buildings, and the work was split up. The engineer designs. Ordering the material is left to the general contractor and its trade partners. When something doesn't fit, it comes back as a request for information, and the answer travels back through the same chain.

I don't blame anyone for that. A building can have 40 or more companies on it, and splitting the work is how you get through that many hands.

But the split has a price. The person drawing the detail rarely sees what it takes to buy it, cut it and put it in. So questions that one engineer could have caught at a desk turn into RFIs on site, with a crew waiting for the answer.

I wish we built more buildings the way we build power plants. I still think the person who draws it should know whether it fits.

From The Punch List (22 Oct): ordering his own rebar on power plants.

To check: "Like your name is on the truck" is our line for "I was really responsible". Does it sound like him?

3The most undervalued person in planning a building is the superintendent.

The most undervalued person in planning a building is the superintendent.

A lot of the superintendents I've worked with came up through the trades. Several told me that in high school the push was toward college, and working construction was looked down on.

They went anyway, and it shows. A good superintendent can look at a plan and tell you what will go wrong when the crew tries to put it in, because they've put it in themselves.

Yet pre-construction is often treated as office work. Estimators price it, project managers schedule it, and the superintendent is often still out finishing the last job. They join when the plan is set, and then they're asked to deliver it.

I understand why. A good super is expensive, scarce and needed on site, and pre-con looks like a bunch of people sitting in a room.

But I'll tell you, for every successful project I ran, I leaned on having a really strong superintendent to help us with pre-con.

That matters more now. People say, let's just ask Claude, and sure, we can. AI can help draft an estimate or a schedule faster than ever. But the experience of actually installing the work is what lets you predict how it will go, and that doesn't come from a model.

So if you run a GC, get your best superintendent into pre-con before the plan is set, and budget that time as part of the job.

From The Punch List (25 Sep): his superintendents, and leaning on them in pre-con.

To check: Is "supers often join after the plan is set" what he saw at DPR?

4With all due respect, Jensen Huang is wrong that China can build a hospital in a weekend. I've built hospitals, and I wouldn't want to be in that one.

With all due respect, Jensen Huang is wrong that China can build a hospital in a weekend. I've built hospitals, and I wouldn't want to be in that one.

He said it a year ago, comparing China with the three years he says it takes to stand up a data center here. People still repeat it.

The bottom line is that China does build faster, and the reason is more interesting than a weekend.

I was talking to Martin Fischer about this. He's a professor at Stanford, leads its Center for Integrated Facility Engineering, and was my advisor. In the early 2000s he went to China to talk about 3D modeling and how to plan buildings in a computer before building them. He went back about eight years later, and they had trained thousands of people who were churning out those models.

Those models opened the door to prefab. When the building exists in detail early in design, you can make big pieces of it in a factory and set them on site. A lot of China's speed came from those eight years of training people in that process.

To be fair to Jensen, his point stands: we're slower. Look at California's high-speed rail, which people call the train to nowhere.

But it's easy to forget how the US won World War II. It out-produced everyone, allies and enemies combined.

If it's a priority, I think the US can be as fast as China, or faster. I'd start where China did, by training people to model buildings early in design.

From The Punch List (22 Oct): his correction of Jensen, and Martin Fischer's China story.

To check: "I've built hospitals" and "I'd start where China did": his words to stand behind?

5I'm not sure I should be saying this as the co-founder of an AI company, but an AI tool that's right 75% of the time can make your construction team slower.

I'm not sure I should be saying this as the co-founder of an AI company, but an AI tool that's right 75% of the time can make your construction team slower.

Here's why. If a tool codes a thousand invoices and gets 750 right, the project manager still has to find the 250 that are wrong, and chasing them can take more time than the tool saved.

So at MeltPlan we hold ourselves to a number we call the 95% threshold. We compare the AI against work a good person has already done well. Above 95%, we think it adds real value. Between 80 and 95%, it's roughly break-even. Below 80%, the person ends up chasing errors, and the tool costs more time than it saves.

The comparison matters too. Compare it to a competent human, not to a perfect world where the stars align. People make mistakes. The PM coding a thousand invoices isn't getting all thousand right either.

To be fair, not every job needs 95%. A rougher tool is fine for a first conceptual estimate you'll redo anyway, but I wouldn't sign a GMP on its numbers.

So when someone shows you an AI tool, send them work your own good person has already done. Run the tool on it, and get the number before anyone on your team relies on it.

From his Beiing Human interview: the 95% threshold.

To check: The 750-of-1,000 example is ours, to show his "below 80" line. Is it fair?

6Owners call it padding. Contractors call it risk. They're talking about the same line in the bid, and they mean two different things.

Owners call it padding. Contractors call it risk. They're talking about the same line in the bid, and they mean two different things.

Padding is money added for no reason, because the contractor thinks it can get away with it. Risk is money held for things that might go wrong and that the contract says the contractor pays for.

I've worked as a GC for a long time, and I hear the padding argument all the time. My issue with it is the margins. Most general contractors I know run somewhere between 1 and 6%, and subcontractors run around 10. One bad price increase can eat all of it.

Every GC has a horror story, and so does every sub. Everyone agrees to be open about the numbers. You show your costs, you sign the deal, and then the prices go up. Once the risk has moved to you, no owner is coming back to pay the difference. That's a fact, and that's business.

So when a contractor holds 10% for unknowns, a lot of it is the price of the risk the owner chose to hand over.

To be fair, some contractors do pad. Owners are right to push for precision, and open books help.

But here's the test I'd love to see more owners run. Tell your contractor: don't do 10% for unknowns, do 2 or 3%, and if it goes beyond that, I'll fund the difference. Then watch what happens to the number. I haven't seen many owners offer it.

From The Punch List (16 Jul): his answer to Jon on "silent padding".

To check: "Some contractors do pad" is our fairness line. Will he say it?

7We planned a project during the first round of tariffs and built it during COVID, and it still came in on budget.

We planned a project during the first round of tariffs and built it during COVID, and it still came in on budget.

My client was Jon Wright, who now hosts The Punch List with Kanav and me. The contract was signed as COVID hit, and the job was physically shut down for six weeks.

Here's what made the budget hold. During planning, we had a lot of conversations, some of them very energetic. Out of them came a set of allowances that were open book, for two things: the tariff risk, and a date the client really wanted to hit.

When COVID shut the job down, the date mattered less to everyone. So we tapped into that open-book bucket to deal with COVID, and the budget didn't change.

To be very honest, we weren't on time. A six-week shutdown added about 10 to 12 weeks. Jon's reaction was, what the hell, why is it more than day for day? Because getting the labor and the resources back takes time. It was one intense conversation.

I'd do the open-book allowance again on any job where prices or rules are moving. It works when the owner and the contractor are both willing to show each other the number.

From The Punch List (16 Jul): his project with Jon through tariffs and COVID.

To check: Fine to name Jon as his client, and would he do the open-book allowance again?

8Bent Flyvbjerg, an Oxford professor, has studied more than 16,000 big projects. Only 0.5% of them came in on budget, on time and with the benefits they promised.

Bent Flyvbjerg, an Oxford professor, has studied more than 16,000 big projects. Only 0.5% of them came in on budget, on time and with the benefits they promised.

That's from his book How Big Things Get Done. Even counting only budget and schedule, fewer than one in ten made both. He calls it the iron law: over budget, over time, under benefits, over and over again.

People outside our industry like to compare a building to a car. You buy a car, and you know the price. His numbers say the building isn't the exception. Nearly every big project goes this way.

His answer is to think slow and act fast. Take your time in planning, where mistakes are cheap, and then build quickly.

From my seat as a GC, construction usually does the opposite. Pre-construction is always rushed, because once a client decides to build, they want the building yesterday. They've done their revenue projections. And pre-con looks like the easiest place to cut time, because it's a bunch of people sitting and talking in a room.

I'll tell you what I've learned about change. During construction, change is bad, and almost every project that goes wrong goes wrong because things changed. During pre-construction, change is good. You beat up your plan several times while it's still on paper, so it can survive what happens on site.

So before you cut weeks out of pre-construction to start sooner, ask your contractor what would have to go right during construction for that to actually save you time.

From his interview (13 Jul), where he cites How Big Things Get Done.

To check: We used the book's numbers (16,000+ projects, 0.5%). He quoted 17,000 and 2% on tape.

9If data centers are more than 40% of your backlog, I'd start winning other work now. I've spent 18 years in construction, most of them at general contractors, and there are two rules I'd hold to if I ran a GC today.

If data centers are more than 40% of your backlog, I'd start winning other work now. I've spent 18 years in construction, most of them at general contractors, and there are two rules I'd hold to if I ran a GC today.

The first is to keep any one market at 30 to 40% of your backlog, 40 at most, and data centers are the market testing that rule right now.

The hard part is the money. If 40% of your work is data centers, I bet that's 70% of your profits. It's a sugar rush.

It's also a natural fit. Most GCs are led by engineers, and a data center is a very engineered building. Nobody complains about how it looks or whether it's comfortable. Every other building is designed for people, and people complain more and ask for more changes. Servers don't. You take care of their cooling, you get them power, and they'll shut up and do their job.

But one market can turn fast. This week Loudoun County in Virginia, the biggest data center market in the country, paused new data center zoning requests for up to a year.

The second rule is to know who you're signing with. A lot of this work is owned through special purpose vehicles. When we signed contracts, we tried to stay away from them if we could, because if they go down, you don't have anyone to collect from.

So if you're over 40%, pick the next two kinds of buildings you want to win and start bidding them this quarter, while the backlog still lets you be choosy.

From The Punch List (30 Oct): the 40% backlog rule, the sugar rush and SPVs.

To check: "A lot of this work is owned through special purpose vehicles" is our framing. Right?

10Everyone is telling California to build homes people can actually buy. You can't do that while nearly every condo project ends up in court.

Everyone is telling California to build homes people can actually buy. You can't do that while nearly every condo project ends up in court.

Condos are 20 to 30% cheaper than other homes, which makes them the obvious first home. In the mid-2000s, about a third of California's new housing was condos. Today it's about 4%.

When we had California Assemblymember Buffy Wicks on The Punch List, she told us that 100% of condo developments in the state go into litigation.

I'll tell you how that looks from a GC's side. Building a condo is such a pain that companies I've worked for won't touch it with a ten-foot pole.

Here's why. Homeowners can sue for construction defects for up to 10 years. So insurance companies assume that anyone building condos is going to get sued, and they price it that way. A GC with condos in its portfolio can pay three to four times more for insurance. That's why most large GCs don't want a condo in their portfolio. Architects carry professional liability insurance too, so many of them stay away as well.

To be fair, zoning and permits matter, and buyers deserve real protection when a building has real defects. Nobody should take that away.

But until the defect rules change, condos will keep losing on the math, whatever we do on zoning. Buffy's condo defect bill died in the last hours of this year's session, and she says she's bringing it back. If you build or buy homes in California, that's the bill to watch next year.

From The Punch List (1 Oct) on condos, with Buffy Wicks's numbers (13 Nov).

To check: "Buyers deserve real protection" is our fairness line on a live bill. Will he take that stance?