Issue #10 (Sunday, Oct 25, 2026)

Good morning, it's Sunday, October 25, and this is The Daily Punchlist. Every super knows the daily report that gets written at 6 p.m. in the truck, from memory, off a phone full of texts. This morning, a Korean builder says an AI agent now does that job for its site managers. Grab the coffee. Here's the punch list.

Today's theme is the gap between using AI and getting something out of it. A Kahua survey of capital program leaders found 84% use general-purpose AI tools, but only 8% report significant measurable value, and bad data is the usual suspect. HD Hyundai Site Solutions will lead a Korean national project to build a digital twin where autonomous excavators can learn on virtual soil before they touch the real thing, and POSCO E&C's company-wide AI challenge produced an agent that saves site managers about 90 minutes a day on daily reports. In Tampa, a self-funded startup 3D-printed the walls of an affordable home in three days, and a union carpenter makes the case that someone still has to know when the AI is wrong. Husqvarna rounds it out with its biggest demolition robots yet.

In today's issue

• Kahua survey: 84% of capital program leaders use AI, only 8% see significant value
• HD Hyundai Site Solutions will lead Korea's digital twin project for autonomous excavators
• POSCO E&C's AI challenge produces an agent that writes the daily report
• A Tampa startup 3D-printed an affordable home's walls in three days
• A union carpenter asks: who will know when the AI is wrong?
• Husqvarna launches its biggest DXR demolition robots

---

1. Kahua survey: 84% of capital program leaders use AI, only 8% see significant value

Kahua released its 2026 Capital Program Maturity Report, commissioned by Kahua and researched independently by Verdantix. It's based on 50 quantitative survey interviews and five in-depth interviews with capital program leaders in the U.S., U.K. and Canada. The headline: 84% of the organizations surveyed use general-purpose AI tools, but only 8% report significant measurable value from their AI investments. Only 4% operate in a fully connected systems environment, and just 16% are very confident that their project and portfolio data is accurate and up to date. 22% have a system of record covering the full project lifecycle, and 58% say they're moving toward fewer systems or a single unified platform over the next few years. Poor data quality, security concerns and difficulty trusting AI outputs were all named as things slowing AI down. "If project information is inconsistent, AI works from those same limitations," said Verdantix analyst Sophie Planken-Bichler.

The fine print: Kahua sells a construction platform for owners that it pitches as a system of record, so the report's prescription matches its product. It's also a small sample of 50 organizations.

Why it matters: Most owners and contractors already have a chatbot open in a browser tab. Very few have project data clean enough for it to do anything useful with. The 76-point gap between using AI and getting value from it is mostly a data problem, not a software one. Garbage in, confident garbage out.

Source: https://www.prnewswire.com/news-releases/only-8-of-capital-program-leaders-report-significant-value-from-ai-investments-new-kahua-research-finds-302913684.html (PR Newswire, Oct 20)

2. HD Hyundai Site Solutions will lead Korea's digital twin project for autonomous excavators

HD Hyundai Site Solutions said on Oct 25 that it has been selected as the lead contractor for a Korean Ministry of Trade, Industry and Energy project to build a high-precision, simulation-based digital twin platform for AI autonomous excavators. The project has a budget of 6.33 billion won and runs until December 2029. The consortium includes Simji Co., the Korea Construction Equipment Research Institute, the Korea Institute of Industrial Technology and the Korea Construction Equipment Industry Association. The idea is to recreate in a virtual space how construction equipment operates and how it physically interacts with soil, then use that platform to design excavators, wheel loaders and articulated dump trucks and to develop and test their autonomy. HD Hyundai Site Solutions will set standard specs for the simulation component database, run field validations and evaluate operating scenarios. In April, the company and Gravis Robotics supplied a 22-ton excavator running their joint Real-X autonomy system to KIBAG's construction site in Switzerland.

Why it matters: The slow part of teaching a machine to dig is that every soil, slope and site is different, and you can't safely test every combination with a real 22-ton machine. A good simulator lets developers run thousands of digs before the bucket hits real ground, then compare the results with real operating data. Dirt doesn't read the manual, so somebody has to model it.

Source: https://www.ajupress.com/view/20261025140870983 (Aju Press, Oct 25)

3. POSCO E&C's AI challenge produces an agent that writes the daily report

POSCO E&C wrapped up a company-wide AI challenge that ran from March 24 to Oct 15, with 1,887 participants, nearly half its workforce, including field staff. The challenge had four categories: video, reports, AI agents and a quiz. The top AI agent entry was a "Work Report Automation AI Agent." Before, partner companies sent work details to site managers through social media messaging, and managers compiled them into daily reports by hand. The agent automates that, cutting a job that took more than 90 minutes a day. POSCO E&C estimates it saves each site manager about 375 hours a year. The company also acknowledged a problem: site data is full of informal language, jargon and typos, which makes it messy to train AI on, and it says scattered site data needs standardizing.

The fine print: the time savings are POSCO E&C's own estimates. The article doesn't say whether the agent is in use on all sites yet.

Why it matters: The daily report is the most universal piece of paperwork in construction, and on most jobs it's still assembled from texts and memory at the end of the day. An agent that reads the subs' messages and drafts it hits a real pain point for every GC. It also lines up with story #1: the hard part isn't the model, it's the slang-filled site data. AI that speaks fluent foreman is the real product.

Source: https://www.ajupress.com/view/20261024112070211 (Aju Press, Oct 24)

4. A Tampa startup 3D-printed an affordable home's walls in three days

The Business Observer profiled Solvenah, a Tampa construction company founded in April (formerly BIA Robotics) that 3D-prints concrete structural walls. In July it printed the 12 wall panels for a roughly 1,300-square-foot affordable home in St. Petersburg's Childs Park neighborhood, developed with the nonprofit Bright Community Trust. Printing took three days and two people, on a $324,000 Build Additive Atlas 3DCP system that prints a 12-by-10-foot wall in about 90 minutes in 17mm layers. Co-founder Carl David says it's the first home in St. Petersburg with 3D-printed walls and that the city had to draft new review procedures for it. Steel reinforcement, insulation and the regular trades are still part of the build. The founders say they've self-funded about $500,000 and plan to move printing into a 3,750-square-foot warehouse on Busch Boulevard, where they can make panels indoors and ship them to sites. An early October tour wasn't flawless: a clog from a test print the day before stopped the Oct. 1 demonstration.

The fine print: the cost and crew comparisons (two people versus a typical crew of 10 or more) are the founders' own figures.

Why it matters: Moving the printer off the site and into a warehouse is the same move precast made decades ago. Weather, mix consistency and moving a $324,000 machine around are what break site printing, and a factory takes care of all three. Fast-setting mortar waits for nobody, including the demo.

Source: https://www.businessobserverfl.com/news/2026/oct/23/tampa-robotics-printed-walls-affordable-neighborhoods/ (Business Observer, Oct 23)

5. A union carpenter asks: who will know when the AI is wrong?

In a Construction Dive op-ed, Lance Mazzariello, a union carpenter and vice president of Specialty Services of New England, argues that construction is adopting AI for estimating, project management, BIM and reality capture without asking who will still have enough know-how to catch the technology's mistakes. He says apprenticeships and AI are both systems for passing on knowledge, and that greater Boston's joint union-employer programs graduate more than 1,000 workers a year. Massachusetts wants to register 100,000 new apprentices statewide by 2036. His prescription is to treat technology as a way to make skilled workers more productive, not to cut headcount, and to share those gains with the workers. "Software can flag a conflict. Someone still has to know how to resolve it," he writes.

Why it matters: Every AI tool in this issue eventually needs a person who can look at the output and say, "That's not going to work." That judgment comes from years in the field, and it doesn't show up in a training dataset. If the industry stops growing it, the tools lose their best checker. The tablet has the drawing. The journeyman has the wall.

Source: https://www.constructiondive.com/news/preserve-trades-knowledge-when-ai-is-wrong/833505/ (Construction Dive, Oct 21)

6. Husqvarna launches its biggest DXR demolition robots

Husqvarna Construction introduced the DXR 705 and DXR 715, the largest class of its remote-controlled DXR demolition robots so far. Husqvarna says both weigh under five tons and several hundred kilos less than demolition robots with similar power. That lets them work in confined, high-risk and hard-to-reach areas. The DXR 705 is built for heavy-duty demolition and industrial work, with heavy breakers, crushers, buckets, grapplers and drum cutters. The DXR 715 has an extended boom for long reach in cleaning, industrial demolition, tunneling and mining. Both can be bought, or leased through Husqvarna's operational lease, which includes insurance and maintenance. The DXR range now has seven models.

The fine print: these are remote-controlled machines with an operator on the controls, not autonomous robots. The power-to-weight claims are Husqvarna's.

Why it matters: Demolition robots are the robots contractors already own. They earn their keep by keeping the operator out of the dust and the danger zone. Packing more breaker into a lighter machine means more jobs where it can fit through the door and the slab can hold it. Big hammer, small footprint.

Source: https://www.europeandemolition.org/communication/husqvarna-introduces-two-powerful-demolition-robots-with-a-compact-design (European Demolition Association, Oct 22)

---

Trending on X

1. @tphuang (a commentator on Chinese EVs, chips and tech), on a 2.5-minute video of construction robots in China, with the same market figures we covered yesterday (CNY18.6 billion in 2025, 28 billion in 2026, 80 billion by 2030): 86 likes, 12 reposts, 2 replies, 3,570 views. The post doesn't name the robots or their makers, so treat the clip as a look at the category, not a product review. https://x.com/tphuang/status/2114151612268638458
2. @LeoKharon (Léo, a robotics research commentator and co-host of @roboticsstack), on RebarSim, an ICRA project that trains a robot entirely in simulation to insert rebar into rack slots: 12 likes, 0 reposts, 4 replies, 656 views. The post reports more than 90% real-world success into a 1.4 mm slot clearance with zero real-world training samples. It's the same simulate-first bet as story #2, just with rebar instead of dirt. https://x.com/LeoKharon/status/2113602377395683372

The Daily Punchline

POSCO E&C says its site messages are full of jargon and typos. Lesson one for the AI: "mud" is not a weather report.

Tool of the day: Punch

With POSCO's daily-report agent in story #3, today's pick from our directory is Punch, a voice-first construction platform. A super tells Punch what happened in the field, and it compiles the daily report, creates purchase orders, assigns punch list tasks and keeps job costs current. It sits between your existing estimating and accounting systems, including QuickBooks, Sage, Foundation and Trimble. Punch says it's built for contractors doing $5 million to $1 billion a year, and it features a drywall superintendent who says he saves 2.5 hours a day on forms. https://dailypunchlist.com/tools/field/punch/ · https://buildwithpunch.com

Funding roundup

• Cross Sector (Machida, Tokyo): Nikon's first spinoff has raised ¥770 million (about $5 million) in total equity and bank loans from seed through Series A, its first funding announcement. Investors include Nikon, Coral Capital, Mizuho Capital, Animal Spirits (which led the Series A), SMBC Venture Capital and Mitsubishi UFJ Capital. Its Frame Finder sensors cut the time to measure a steel column during erection from 15 minutes to 15 seconds, the company says, and it has run trials or test deployments on more than 30 sites. The money goes to ramping up sensor production and developing an automatic steel adjustment mechanism (Oct 22). https://prtimes.jp/main/html/rd/p/000000004.000192947.html