Tech & AI Roundup — September 8, 2026

Samwise Tech/AI/Robotics Newsletter

Tuesday, September 8, 2026

AI  ·  Robotics  ·  Hardware  ·  Research  ·  Regulation
Tech, AI, and robotics news, carefully curated and summarized daily
AIINDUSTRY

Eric Wu exits stealth with NavigateAI, an AI copilot for construction

Eric Wu, who built and ran Opendoor before stepping away in 2022, has launched NavigateAI, an AI copilot for construction workers. The product runs on smartphones and Meta AI glasses, giving workers hands-free, real-time guidance on build specs and code compliance. NavigateAI came out of stealth in May 2026 with $25 million at a $225 million valuation, backed by Elad Gil, Khosla Ventures, Fifth Wall, and Lennar. The company targets a construction labor shortage so severe that data center projects now require up to 6,400 workers apiece. Longer term, Wu says the egocentric video dataset of workers building correctly and incorrectly will prove as valuable as the software business itself.

ROBOTICSINDUSTRY

Agility Robotics reveals $1.8M revenue against $140M loss ahead of $2.5B SPAC

Agility Robotics has filed an S-4 with the SEC ahead of a planned SPAC merger with Churchill Capital Corp. XI, revealing its financials for the first time. The filing shows $1.8 million in 2025 net sales against a $140 million operating loss—a roughly 1,400× valuation-to-revenue multiple at the deal’s $2.5 billion price. The SPAC is expected to raise more than $620 million, including a $200 million PIPE led by Foxconn. Agility’s Digit humanoid has logged more than 65,000 operating hours across 9 customer sites, with one customer holding $300 million in multi-year Digit v5 orders. The company projects 800 deployed units by 2027, scaling to 25,000 by 2035.

AIINDUSTRY

Enterprise AI spending surges but almost no company can prove ROI

When Uber handed its engineers Claude Code and set up token-consumption leaderboards in December 2025, the entire 2026 AI coding budget was gone by April—with no proven link, per Uber COO Andrew Macdonald, to shipping better products. The episode coined a term: “tokenmaxxing”—surging AI token consumption without measurable ROI. Gartner projects enterprise AI agent software spending at $207 billion in 2026, up 139% year-over-year, but token costs don’t model like traditional software. Companies are now deploying smart routing, model tiering, and spend caps. Of all enterprises VentureBeat spoke with, only Everlaw could share hard ROI figures: $3,500 in tokens saved seven engineer-months on a core infrastructure overhaul.

ROBOTICSREGULATION

MassRobotics survey finds industry evenly split on FCC foreign-robot import ban

MassRobotics surveyed 14 member companies—from solo-founder teams to organizations with 200-plus employees—on the FCC’s July 2026 ban on new imports of foreign-produced mobile robots and humanoids. The results were an even split: 43% called the ruling “beneficial” or “very beneficial,” 43% called it “detrimental” or “very detrimental,” and 14% expected no impact. Sentiment tracked closely with manufacturing footprint—domestic producers viewed the ruling as a competitive advantage, while those with foreign-dependent supply chains saw it as costly disruption. The FCC rule covers connected robots weighing over 4.4 lb., with a 65% domestic-component threshold, rising to 75% by 2029. The single most-requested form of support: vetted onshoring supply networks.

ROBOTICSHARDWARE

FDA clears Vitestro Aletta: the first autonomous blood-draw robot for non-hospital use

Vitestro’s Aletta is the first autonomous blood-draw device authorized by the U.S. FDA for non-hospital settings. The Dutch medical robotics firm’s machine uses near-infrared imaging, ultrasound, and Doppler flow mapping to locate veins, position a needle, and collect samples with no human contact. In a clinical trial of more than 1,600 people in the Netherlands, Aletta achieved a 94.5% first-attempt success rate, including among patients with obesity and difficult veins. One phlebotomist can supervise up to three machines simultaneously—a capability Vitestro says will address the roughly 25% annual turnover and 10% vacancy rate in phlebotomy. European commercial launch is planned for next year, followed by the U.S.

ROBOTICSRESEARCH

Inbolt CEO: physical AI has a deployment problem, not a data problem

Physical AI’s consensus: progress is gated by data—more, better curated, better simulated. Inbolt CEO Rudy Cohen says that logic fails in the factory. After deploying vision-guided control across more than 100 factories and 40 million-plus robot cycles, Cohen argues the real bottleneck is the perception-to-motion loop and the integration overhead that makes every deployment a six-figure project. At RoboBusiness 2026 in Santa Clara on October 20–21, Cohen will draw on production data from Stellantis, Toyota, and Ford to argue that physical AI is already commercially proven—just not where headlines are looking. Inbolt has raised €20 million since its 2019 founding and operates in Europe, the U.S., and Japan.

AIRESEARCH

TechCrunch AI glossary adds “opaque recurrence”—the reasoning technique alarming safety researchers

TechCrunch has updated its ongoing AI vocabulary glossary—written by Natasha Lomas, Romain Dillet, Kyle Wiggers, and Lucas Ropek—adding entries for terms entering the field alongside increasingly complex frontier models. The headline new entry is “opaque recurrence”: the reasoning technique used in OpenAI’s new Astra model that has AI safety researchers alarmed. Rather than generating visible step-by-step chain-of-thought, an opaque-recurrence model routes a query through its internal layers in a loop—more compute-efficient but producing fewer readable traces, raising interpretability and safety concerns. The glossary is a living document, updated regularly as vocabulary in the field evolves.

⚡ Tech Pulse

FRONTIER MODELS:
GPT-6 Astra (OpenAI) — GPQA 96.0% (#1) | Claude Fable 5.1 (Anthropic) — #2 | Claude Opus 5 (Anthropic) — #3
OPEN WEIGHT:
Kimi K3 (Moonshot AI) — GPQA 93.5% (#1 Open)
SMALL & EDGE:
Qwen 3.8 (Alibaba) — LLM Stats 52.1, $1.81/M tok | Gemini 3.8 Flash (Google) — LLM Stats 51.0, 239 tok/s, $0.89/M
MARKET CAPS (AI):
Nvidia $5.58T • Alphabet $4.10T • Microsoft $3.71T
MARKET CAPS (ROBOTICS):
Intuitive Surgical $175B • ABB $165B • Symbotic $33B
Browse All Past Newsletters

Every newsletter preserved and searchable