15 Technologies — Week of May 18, 2026
Pre-trained AI models that give robots the ability to perceive and manipulate physical objects in the real world, freely available for anyone to use and modify. Think of them as a "base brain" that can be fine-tuned for specific tasks — like ChatGPT for robots.
Warehouse automation, manufacturing assembly lines, surgical robots, agricultural harvesting, construction site robots. Ai2's MolmoAct 2 (May 2026) runs 37x faster than its predecessor and handles real-world manipulation tasks that previously required custom engineering for each scenario.
Connects to: Multi-Agent Orchestration (robots as agents in factory workflows), Edge AI (on-device inference without cloud), Physical AI Simulation (training before deployment).
Day 1-2: Watch the MolmoAct 2 demo videos at allenai.org to understand current robot capabilities. Day 3-5: Browse the open-weight model on HuggingFace — see what data inputs the model takes and what outputs it produces. Day 6-7: Map one PGC field task (e.g., glass panel handling, material staging) to where a robotic arm could assist — no purchase needed, just research.
Open robotics models mean hardware companies can build capable robots without reinventing the AI brain — expect a wave of affordable physical automation in the next 18 months.
Threat actors using large language models to autonomously discover previously unknown software vulnerabilities and build working exploits — no human expert required. Google GTIG confirmed the first known AI-developed zero-day in May 2026: an AI found and weaponized a 2FA bypass in an open-source admin tool.
Industrial-scale cyber attacks, ransomware campaigns targeting construction firms (often SMBs have weak security), supply chain attacks on tools like project management software. Criminal groups are on the verge of using AI to conduct mass exploitation campaigns automatically.
Connects to: Agentic AI (same models, different application), Privacy-Preserving AI (on-device inference reduces attack surface), Agent Memory (AI persisting and learning attack patterns).
Day 1: Read Google's GTIG May 2026 report on AI vulnerability exploitation (free, cloud.google.com/blog). Day 2: Audit PGC's most critical software: what vendors have known vulnerabilities? Use exploit-db.com or cve.mitre.org to check. Day 3: Enable two-factor authentication on the single most critical account (likely the primary estimating or project management tool) — this blocks the confirmed zero-day bypass method.
The economics of cyber crime have shifted: AI lets attackers find vulnerabilities faster than human developers can patch them. Every month you delay basic hygiene (2FA, patches) widens your attack surface.
Next-generation EV batteries that replace expensive lithium with abundant sodium — cheaper, safer, works in freezing temperatures. CATL and Changan launched the Nevo A06 in February 2026, the world's first mass-produced sodium-ion EV. CATL's Naxtra batteries now operate from -40°C to 70°C with energy density up to 175 Wh/kg.
Electric work vehicles, site equipment (forklifts, lifts), grid energy storage for buildings, cold-climate operations. Sodium-ion batteries cost less to manufacture and source, which should drive down electric work vehicle prices significantly by 2027-2028.
Connects to: Edge AI (battery management systems getting smarter), IoT Sensors (more sensors running longer on smaller batteries), Autonomous EVs (the vehicles that will use these batteries).
Day 1: Research sodium-ion availability via CATL Naxtra and Changan Nevo A06. Day 2: Identify one PGC work vehicle that could go electric — a site buggy, small delivery van, or forklift. Check available models and charging infrastructure needed. Day 3: Get a quote from a commercial EV dealer for an electric light truck — just to understand pricing and lead times. No commitment needed.
Sodium-ion breaks the lithium supply chain dependency that has kept EV prices high — construction fleets may go electric faster than expected, and battery-powered site equipment will get cheaper.
Frameworks that coordinate multiple AI agents working together on complex tasks — one agent writes code, another reviews it, a third fixes errors, all orchestrated by a "manager" agent. Tools like Composio's Agent Orchestrator let parallel coding agents each work in their own git worktree, handling CI fixes, merge conflicts, and code reviews autonomously.
Enterprise workflow automation, complex coding projects, multi-step business processes (quote → review → send → track). Anthropic just released "Dreaming," "Outcomes," and "Multi-Agent Orchestration" for Claude Managed Agents. Microsoft Copilot Studio now orchestrates agents across apps and data.
Connects to: MCP Protocol (standardized tool connections for agents), Agent Memory (shared context across agents), Open-Source Robotics (multi-agent coordination for factory floors).
Day 1: Watch the Composio Agent Orchestrator demo on GitHub (free, github.com/ComposioHQ/agent-orchestrator). Day 2-3: Identify a PGC workflow that involves multiple steps across different systems — like "field report → data entry → alert PM." Map the handoff points. Day 4-7: Prototype a simple version using a no-code automation tool like Zapier or n8n — even if it only covers one step, you'll see the pattern.
Single AI agents handle one task; multi-agent systems handle whole workflows — this is what turns AI from a novelty into a competitive advantage for back-office operations.
A universal standard (now under the Linux Foundation's Agentic AI Foundation) that lets AI models connect to external tools, databases, and data sources through one standardized interface — instead of building custom integrations for every AI × every tool combination. As of May 2026: 97 million monthly SDK downloads, 9,400+ public servers, native support from Anthropic, OpenAI, Google, and Microsoft.
Connecting AI assistants to your company's data (Drive, calendar, CRM), enabling AI agents to use real tools (send emails, query databases, call APIs), building custom AI integrations without starting from scratch.
Connects to: Multi-Agent Orchestration (MCP as the "wiring" between agents), Agentic AI (agents need MCP to act on real-world data), Agent Memory (MCP enables persistent context).
Day 1: Read "What is MCP" primer on dev.to (search: "Model Context Protocol 2026 primer"). Day 2: Go to smithery.ai — a marketplace for MCP servers — and find one that connects to Google Drive or Gmail. Day 3: If you use Claude or Cursor, enable an MCP server in 10 minutes (free tier available). See how it pulls real data from your connected accounts.
MCP is the USB-C of AI integration — it standardizes how AI connects to everything else, meaning custom integrations that used to take weeks now take hours.
Brain-inspired computer processors that process and store information simultaneously — like neurons — using 70% less energy than traditional chips. As of February 2026, neuromorphic computers can now solve complex physics simulation equations that previously required energy-hungry supercomputers. A new nanoelectronic device using modified hafnium oxide mimics how neurons process information.
Real-time structural analysis (simulating wind loads on buildings), climate modeling, drug discovery, on-device AI inference for autonomous vehicles, edge computing in remote locations. The energy efficiency means complex AI can run anywhere.
Connects to: Edge AI (neuromorphic chips power real-time AI without cloud), Physical AI Simulation (simulation runs faster and cheaper), Text-to-CAD (faster physics sim for structural validation).
Day 1: Watch the ScienceDaily video on neuromorphic physics simulation (free, sciencedaily.com). Day 2: Look up Intel Loihi or IBM NorthPole — current neuromorphic chips — and their benchmark results. Day 3: Think about where PGC runs physics-heavy calculations: structural glazing load calculations, facade thermal modeling. Could faster simulation change how you bid complex projects?
Neuromorphic chips bring supercomputer-level physics simulation to edge devices — in 3-5 years, a tablet on a job site could run thermal and structural models in real time.
Engineered materials that detect and repair internal damage autonomously — mimicking biological healing. The global market hit $4.99 billion in 2026, growing 26.3% from 2025. Combined with IoT sensors embedded in structures, these materials enable buildings that monitor their own health and self-repair cracks before they spread.
Concrete infrastructure (bridges, parking structures), glass sealants and coatings, facade panels, high-rise curtain walls. Self-healing concrete uses bacteria or chemical agents activated by water ingress. Smart coatings on glass can detect stress fractures and seal them automatically.
Connects to: IoT Sensors (embedded sensors trigger healing), Computer Vision (structural health monitoring), Edge AI (on-device processing of sensor data).
Day 1: Read the Self-Healing Materials Research Report 2026 (globenewswire.com, free). Day 2: Look up BASF or HeidelbergCement self-healing concrete products — note pricing and application methods. Day 3: Identify one PGC project where self-healing sealant could reduce callback costs — large glass facade in coastal environment? Request a sample from a supplier.
The biggest cost in glazing isn't the glass — it's callbacks for sealant failures and water leaks. Self-healing materials starting to reach commercial availability could virtually eliminate that problem within a decade.
AI systems that automatically compare photos of a construction site against BIM models or schedules to track progress, flag delays, and document conditions. Tools like OpenSpace capture 360° imagery on walk-throughs and automatically map what exists to what was planned — tracking 700+ components across trades without BIM files needed.
Weekly progress tracking for PMs, quality control documentation, subcontractor coordination, owner progress reports, punch list generation. TrueLook and similar systems generate time-lapse documentation and integrate with Procore and Autodesk.
Connects to: 3D Scanning / LiDAR (complementary capture methods), IoT Sensors (sensor fusion for fuller picture), Agentic AI (agents that act on CV output — notify PM, generate report).
Day 1: Watch OpenSpace's product demo (openspace.ai, free trial). Day 2: Download the OpenSpace app — it's free for one project. Walk a current PGC job site and capture 360° photos. Day 3: See the automatic progress comparison — how does it match against your schedule? Evaluate whether the time savings on documentation justifies the subscription cost.
Every PM loses 3-5 hours/week documenting progress. CV-based progress tracking automates that work and gives owners real-time visibility — a competitive differentiator on large projects.
iPhones and iPads since iPhone 12 Pro have LiDAR sensors that can capture millions of 3D data points in minutes — creating accurate room scans, site maps, and point clouds. Apps like SiteScape, Canvas, MagicPlan, and Scanbrix turn these scans into professional CAD files (SketchUp, Revit, DWG) on demand.
Existing conditions surveys (replacing measuring tapes), prefabrication measurement verification, as-built documentation for punch lists, change order documentation with 3D evidence, glass ordering reference scans for complex openings.
Connects to: Computer Vision (scan + CV = automated progress tracking), Text-to-CAD (scan data → CAD model), Self-Healing Materials (scan before/after to document repairs).
Day 1: Download Scanbrix (free scan, pay for CAD export) or MagicPlan (free tier). Day 2: Use your iPhone (Pro model) or iPad to scan one complex PGC opening — a tough corner or unusual angle. Export the floor plan. Day 3: Compare it to your manual measurement — how close? Evaluate for use in preliminary measuring before full fieldwork.
You're already carrying a $1,000+ laser scanner in your pocket. The accuracy is good enough for preliminary takeoffs and change order documentation today — and improving fast.
AI systems that autonomously write, review, test, and deploy code — handling entire feature development from spec to production. These go beyond autocomplete: they own a task end-to-end, run tests, fix errors, and open pull requests. The enterprise shift in 2026 is moving from "AI assist" to "AI engineer" — autonomous agents managing codebases.
Internal tool development, automating legacy code modernization, QA test generation, DevOps pipeline creation. A single developer + AI coding agent can now do what used to require 3-4 engineers on routine tasks.
Connects to: Multi-Agent Orchestration (multiple AI coding agents working together), MCP Protocol (connecting agents to internal tools), Agent Memory (agents remembering codebase context).
Day 1: Try Cursor (free tier) or GitHub Copilot on a small PGC internal tool task — even a simple script. Day 2: Give it a multi-step task: "Write a script that parses these measurement files and generates a summary." See how far it gets autonomously. Day 3: Evaluate: what did it get right? What needed human correction? Where does it save time?
AI coding agents mean small teams can build more internal tools — automating the work that used to require a dedicated developer. PGC could have custom software built for specific workflows without hiring engineers.
headsets like Microsoft HoloLens 2 and Apple Vision Pro overlay digital information onto the physical world — showing workers step-by-step instructions, remote expert annotations, or BIM models on the actual installation space. Industrial smart glasses have moved from pilot phase to mainstream deployment in 2026, with RealWear Navigator Z1 and HoloLens 2 leading field service and training use cases.
Installation training for complex curtain wall systems, remote expert support (expert draws annotations in worker's field of view), maintenance overlay showing internal schematics, quality inspection overlays comparing installed work to BIM specs.
Connects to: Computer Vision (AR uses CV to understand the space), IoT Sensors (sensor data displayed in AR), AI Coding Agents (generate the AR content automatically from BIM/manufacturing data).
Day 1: Research HoloLens 2 pricing (~$3,500) and rental options for enterprise. Day 2: Watch industrial use case videos on YouTube — search "HoloLens 2 construction field service." Day 3: Identify one PGC installation task where AR could reduce training time or callbacks — complex curtain wall panel? Unusual fixing detail? Rent a device for one week on a trial project.
AR glasses let a senior glazier mentor a crew anywhere in the world without traveling — and give new workers hands-free guidance on complex installs. The ROI is real for tasks with high training cost or callback risk.
Running AI models directly on devices (phones, tablets, laptops, IoT hardware) instead of sending data to cloud servers. The performance leap in 2026 means complex AI — voice transcription, image recognition, code generation — runs locally without internet. This addresses both latency and data privacy concerns that have held back field AI adoption.
Field data capture with AI analysis (photo → report instantly, no upload needed), offline AI tools on remote job sites, privacy-sensitive data processing (health records, financial data never leaves the device), real-time defect detection on glass lines.
Connects to: Neuromorphic Computing (hardware enabling edge AI), IoT Sensors (edge AI processes sensor data locally), Computer Vision (CV models running on site cameras locally).
Day 1: Download LLM apps that run locally — like Octane AI or LM Studio — on your work laptop. Try an open-source model like Llama 3.2. Day 2: Test offline: turn off WiFi and ask it to summarize a field report, generate a follow-up email. Day 3: Evaluate: does it handle your most common field task prompts adequately? Could an iPad + local AI replace a cloud subscription for field use?
Edge AI means AI works in basements, tunnels, and remote sites with no connectivity — and your data never leaves the device. For construction companies with spotty site connectivity, this is a game changer.
AI agents that can hold money, make payments, negotiate, and transact — autonomously spending and earning on behalf of humans and businesses. The critical unlock in 2026: reliable, compliant payment infrastructure that lets agents make real transactions (not just promises). This is moving from demos to actual commerce — agents buying and selling services, paying contractors, billing clients.
Automated vendor payments, real-time subcontractor settlements, AI-managed project budgets that pay for resources as needed, instant change order processing, automated permit fee payments, AI assistants that buy things on your behalf.
Connects to: Multi-Agent Orchestration (agents coordinating payments across workflows), Agentic AI (agents with real money at stake), MCP Protocol (agents connecting to banking/financial APIs).
Day 1: Research "agentic AI payments" on TechCrunch or VentureBeat — search for recent articles on AI-to-AI payments. Day 2: Look at Stripe Agent Toolkit or Plaid — these are the infrastructure being built right now. Day 3: Identify one PGC payment workflow that could be automated: material vendor payment when delivery confirmed? Retainage release when sign-off received? Map the trigger and the payee.
When AI agents can transact money autonomously, the latency between work done and payment received drops to zero. Subcontractors could get paid same day — eliminating the cash flow anxiety that plagues construction.
AI systems that convert natural language descriptions or sketches into detailed CAD models, BIM files, or parametric 3D geometry — no CAD software expertise required. Recent versions generate not just geometry but also specification data, material schedules, and structural load information.
Quick conceptual studies from client descriptions, generating multiple facade options for client review, automating repetitive CAD tasks like window schedule generation, converting scanned sketches to parametric models for fabrication.
Connects to: 3D Scanning (scan → AI understands existing conditions → generates design), Physical AI Simulation (AI validates generated design against structural requirements), Edge AI (on-site design iteration).
Day 1: Try Tripo3D or Meshy.ai — free tiers exist for text-to-3D. Describe a simple glass panel detail and see what it generates. Day 2: Try the SketchUp AI beta or similar tools emerging in 2026 — search for "AI to SketchUp" on YouTube. Day 3: Take one PGC detail drawing and see if current AI tools can generate a parametric version — even partially. Identify where human correction is still needed.
Text-to-CAD won't replace CAD designers for 5-10 years, but it will automate the 40% of repetitive work that designers hate — letting them do higher-value concept development instead.
AI systems that maintain persistent memory across sessions — remembering your preferences, past conversations, project history, and accumulated context — combined with the ability to reason over very long documents (100K+ tokens). The emerging pattern: agents that learn from every interaction and get better at your specific needs over time.
Personal AI assistants that know your business deeply (remembering past bids, client preferences, project quirks), long-document analysis (entire project specs, contract documents), cross-session continuity (picking up where you left off without re-explaining context), institutional knowledge retention.
Connects to: MCP Protocol (memory connects to external data sources), Multi-Agent Orchestration (agents share memory across a team), Agentic AI (agents that learn and improve over time).
Day 1: In your AI assistant (Claude, ChatGPT, etc.), start a project-specific thread and reference it a week later — see if it remembers context. Day 2: Try uploading a long document (100+ page spec) to an AI and asking cross-referencing questions — test the long-context reasoning. Day 3: Evaluate: for PGC, what repetitive context do you re-explain every session? Can you build a persistent context file the AI reads each time?
Today's AI "forgets" everything when a session ends. Agent memory means your AI gets genuinely smarter about your business over time — less re-explaining, more acting on accumulated understanding.
| # | Technology | Category | Glazing Applicability | MVP Difficulty |
|---|---|---|---|---|
| 1 | Open-Source Robotics Foundation Models | Robotics | Medium — 3-5 yr horizon | Research only |
| 2 | AI-Developed Zero-Day Exploits | Security | High — applies now | Low — do it today |
| 3 | Sodium-Ion Battery Mass Production | Energy | Medium — fleet transition | Medium — investigate |
| 4 | Multi-Agent Orchestration | AI | High — back-office automation | Medium — prototype |
| 5 | Model Context Protocol (MCP) | Dev Tools | High — immediate value | Low — try today |
| 6 | Neuromorphic Computing for Physics Sim | Computing | Medium — structural analysis | Research only |
| 7 | Self-Healing Materials + IoT | Materials | High — sealant/cladding | Medium — evaluate |
| 8 | Computer Vision Progress Tracking | Construction Tech | High — PM documentation | Low — try free trial |
| 9 | LiDAR 3D Scanning (Mobile) | Field Tools | High — measurement | Low — use iPhone |
| 10 | AI Coding Agents | Dev Tools | High — internal tools | Medium — evaluate |
| 11 | AR / Mixed Reality Field Ops | AR/VR | Medium — training/remote expert | High — device cost |
| 12 | Edge AI on Local Devices | AI | High — site connectivity | Low — try offline AI |
| 13 | AI Agents as Payment Rails | FinTech | Medium — cash flow | High — infrastructure |
| 14 | Text-to-CAD / AI Design Gen | Design | High — concept development | Low — try free tools |
| 15 | Agent Memory + Long-Context | AI | High — all AI interactions | Low — configure today |