2026-10-19

Issue #4 (Monday, Oct 19, 2026)

馃帶 Listen to today's episode 路 6 min

Get this in your inbox every morning.

Good morning, it's Monday, October 19, and this is The Daily Punchlist. Happy Monday, and today belongs to the tunnels. Two Japanese contractors announced big steps today: Obayashi finished field-testing all six parts of its system to automate shield tunneling, and HAZAMA ANDO, NTT and 1Finity showed engineers could inspect a mountain tunnel's walls for 0.2 mm cracks over a link simulating 1,000 km. Closer to the office, a roofing contractor showed the AI agents that now handle his invoices, supplements and material orders, and an audit firm opened up its AI-assisted payment-application checks to project owners. Plus, a Danish startup's robot builds concrete walls with no formwork, and AI rendering moves inside Revit, SketchUp and Rhino.

In today's issue

1. Obayashi finishes field-testing all six parts of its automated tunneling system

Obayashi said it has completed individual field trials of all six technologies that make up OGENTS (Obayashi Intelligent Shield), its system for automating shield tunneling. That's the method where a cylindrical machine bores through the ground while curved "segments" are assembled into rings behind it to form the tunnel wall. The six pieces are automatic surveying, automatic steering of the shield machine, automatic removal of excavated soil, automatic backfill grouting, automated shield equipment, and automatic segment assembly. They use AI, machine learning and advanced sensors to make some of the technical calls operators used to make from experience, which Obayashi says keeps decisions more consistent. The trials ran at 10 shield tunnel sites in Japan starting in March 2021. Surveying took one person instead of two without losing accuracy, automated steering held the tunnel alignment while easing the operator's workload, and automated segment assembly let one operator build rings precisely without working at the dangerous tunnel face. Based on those trials, Obayashi estimates that using all six together on a job would need roughly 40% to 50% fewer workers than a conventional crew. By December 2027 it plans to put every piece on a single site for an integrated test.

Why it matters: Automating one task rarely cuts a crew by much, which is why Obayashi is linking all six. The 40% to 50% figure is an estimate from separate trials, so the 2027 combined test is the number to watch.

Source: https://www.obayashi.co.jp/news/detail/news20261019_1.html (Japanese, Oct 19)

HAZAMA ANDO, NTT and 1Finity (a Fujitsu company) announced a completed proof of concept for running tunnel inspections remotely over NTT's IOWN all-photonics network. The test, run in March at HAZAMA ANDO's Tsubakihara Tunnel project on Japan's Tokai-Hokuriku Expressway, laid 1,100 m of fiber inside the tunnel with six optical network nodes every 100 m from the tunnel face. The gear sat in IP65-rated enclosures and kept working through dust, vibration and spring water. The companies report no packet loss on the fiber section and more than 24 hours of continuous operation. The network moved about 20 GB of 3D scan data to the site office in roughly one minute while also carrying video from the equivalent of 15 cameras. Over a link simulating a remote office 1,000 km away, inspectors could see 0.2 mm cracks (one of the tunnel-wall acceptance criteria) on 4K video, and NTT's AI image analysis detected the cracks. Next come trials on a live IOWN network linking several tunnel sites, remote offices and data centers, with phased use on real tunnels starting in fiscal 2027.

Why it matters: Tunnel faces are dangerous and remote, and skilled inspectors are scarce. If one specialist can check several tunnels from a city office, and AI flags cracks first, the bottleneck shifts from travel to bandwidth.

Source: https://group.ntt/en/newsrelease/2026/10/19/261019a.html (NTT, Oct 19) 路 Full release (PDF): https://group.ntt/en/newsrelease/2026/10/19/pdf/261019aa.pdf

3. A roofer shows the AI agents that now run his back office

Roofing CRM maker Contractors Cloud published a recap of its first "AI Office Hours." In it, Adam Lynn of First National Roofing, a roofer for 17 years who says "I am not a tech guy," showed the agents he runs through the company's MCP (Model Context Protocol) server. One agent has its own email address: it splits the supplier's batch invoice PDFs, files each invoice to the right project and, as of last week, enters the expenses. That used to take an office associate about an hour per batch. Other agents upload EagleView reports, update projects from public adjuster emails, set up a project when a contract is signed, send a morning report on yesterday's paid leads, and draft supplements from the EagleView, the adjuster's estimate and Xactimate pricing. They also draft material orders using his ordering standards. Lynn says nobody was let go, a person still reviews the work, and he has plans for 21 agent "employees." In a live demo, Contractors Cloud's Mike Kampa ran a whole job from Claude, starting from a State Farm scope-of-loss PDF and an EagleView PDF. It created the project, matched the insurance scope and flagged about $1,700 of non-trade items, and generated material and crew work orders using local Minnesota code rules. Asked "What am I missing on this job?", it caught a missing deductible on the invoice. Agents work with each user's existing app permissions. Hover and ABC Supply integrations are next.

Why it matters: This is a small contractor, not a big GC with an innovation team, using off-the-shelf AI on top of a CRM. His advice is the useful part: start with the repetitive task that eats an hour of someone's day, and give each agent one job. It's the vendor's own recap, so treat the details as their account.

Source: https://contractorscloud.com/blog/ai-office-hours-recap-how-first-national-roofing-is-automating-with-the-contractors-cloud-mcp-server/ (Oct 17)

4. ERLEtek's robot builds concrete walls layer by layer, no formwork

ERLEtek, a seven-person spinout of the Technical University of Denmark founded in 2025, is building robots that put up retaining walls and flood-protection walls without formwork. The robot fits in the back of a van. On site it uses sensors and GPS to find its position, downloads the plans and lays concrete one layer at a time. Today it builds walls 10 to 40 cm thick, and a version in development reaches 2 m tall. Its 500-liter hopper holds about half an hour of material and can be refilled by the dumpers already on site. It works with standard concrete, lower-CO2 concrete and clay-based mixes, and a sensor tracks the material going in so the software can adjust vibration and flow. ERLEtek says a 100 m wall that takes two workers about 280 labor hours takes the robot about 80 hours. That is the company's figure. At the end of October it plans to print a 10 m long, 1 m high wall in a test pool, flood it and measure the water pressure, and next summer it plans a 100 m coastal-protection wall on a Danish island.

Why it matters: Skipping formwork is where ERLEtek's claimed labor savings come from. Long, low walls like flood defenses and retaining walls are a practical first job for printing robots, and the flood test later this month will show whether the walls hold back water.

Source: https://ioplus.nl/en/posts/housing-needs-infrastructure-can-robots-build-it-faster (IO+, Oct 19)

5. Construction auditors open their AI payment-application checker to owners

Fort Hill Associates, a Greenville, South Carolina construction audit firm that has worked on more than 1,600 capital projects since 2006, launched Fort Hill LogiX as a standalone company so project owners can use its engine directly. AI reads the documents, and fixed rules test each line of a payment application against the owner's contract, including CMAR and GMP agreements, labor rates and change orders. Every exception comes back with the page it was found on. A Certified Construction Auditor works through each exception and signs off the report before the owner sees it, and the owner makes the final call. Fort Hill Associates built LogiX, is its first customer and says it stands behind every report. The firm says the work follows the National Association of Construction Auditors' professional standards.

Why it matters: Fort Hill says the hours needed to check every line by hand outgrew what any audit practice could staff. Pairing AI that checks every line, with the page attached, with a credentialed auditor who still signs is a sensible split. If owners adopt it, GCs should expect more line-item questions on their pay apps.

Source: https://www.prnewswire.com/news-releases/the-auditors-who-got-tired-of-auditing-by-hand-built-a-machine-now-owners-can-use-it-302910555.html (Oct 15)

6. mnml puts AI rendering inside Revit, SketchUp and Rhino

mnml.ai launched free plugins for Revit 2024 to 2027, SketchUp 2026, Rhino 8 and Grasshopper. Designers can render the view they're working on from a panel beside the model, with no export or upload. A render can go back into the SketchUp or Rhino model, or into a Revit drafting view, in one step. In Grasshopper, 2 to 8 design options render together in one batch. The plugins are free, but renders use mnml credits: 25 credits per render on the default engine, with the price shown on each button before you click. Free accounts can render with a daily limit and watermarked images. The Revit plugin needs full Revit (Revit LT doesn't load add-ins). mnml says it doesn't train AI models on customer work and that teams at more than 2,400 firms use it. Those are company figures.

Why it matters: The export-upload-download loop was the main friction in AI rendering. Inside the model, testing a material or massing option becomes a quick check rather than a side project. Watch your credit spend if a whole team adopts it.

Source: https://mnml.ai/press/mnml-launches-plugins-rhino-grasshopper-sketchup-revit (Oct 18)

Trending on X

  1. @andreasklinger, an investor's video visit to AUAR in London, which builds container-sized "micro factories" (a robotic arm, a robotic saw and a robotic assembly table) meant to be shipped to a site to build wall panels there: 73 likes, 6 reposts, 8 replies, 2 quotes (2,648 views, 19 bookmarks). Klinger discloses he's a small investor. He says AUAR is finishing a micro factory for a US East Coast home builder for 40 homes. His headline claim, that 200 of these could cover the UK's yearly housing need, is his own and not verified. https://x.com/andreasklinger/status/2112096741491155292
  2. @WOLF_Financial, a list attributed to J.P. Morgan that ranks where humanoid robots will work, from easiest to hardest: 23 likes, 5 reposts, 8 replies, 0 quotes (11,363 views). Construction lands in the middle tier, "working with trained adults," alongside agriculture and reverse logistics. That makes it harder than factories but easier than homes and retail. We haven't seen the underlying report. https://x.com/WOLF_Financial/status/2111904441376932279
  3. @0xForem, a six-second clip that appears to show a two-legged robot carrying a concrete block across a site and setting it in place: 22 likes, 0 reposts, 4 replies, 0 quotes (1,054 views). Heads up: this video is highly suspected to be AI-generated, and its content is not verified. The post names no company, site or date. https://x.com/0xForem/status/2111819591831957900

Tool of the day: Cupix

If remote tunnel inspection (#2) got you thinking about checking sites without visiting them, Cupix is the reality-capture pick in our directory. It turns 360掳 video walks into measurable 3D models over time (a "4D" record), so teams can look around, measure, compare work against BIM and track changes remotely. It brings laser scans, drone imagery and BIM into one place, and it integrates with Autodesk, Procore and Revizto. The company says 70% of ENR's Top 30 general contractors use it. That's a company figure. https://dailypunchlist.com/tools/reality-capture/cupix/ 路 https://www.cupix.com

Funding roundup

A quiet weekend for deals. We couldn't confirm any new AI-in-construction rounds announced since Friday, so here are two from last week we hadn't covered yet.

Get this in your inbox every morning.