Google I/O 2026 was not a single product announcement. It was a map of where Google thinks AI work is heading. The headline items were easy to list: Gemini 3.5 Flash, Gemini Spark, Gemini Omni, Android XR smart glasses, Antigravity 2.0, AI Mode updates and a $100 per month AI Ultra tier. The more interesting point is how these pieces fit together for industries that do not live inside a chat window.
For AEC, the most relevant shift is the movement from AI as a destination to AI as a layer across devices, workflows and context. The workbook inputs rightly call out four practical hooks: Omni for video and visual workflows, XR glasses for site inspection, Spark for project administration and Antigravity 2.0 for construction technology teams building agents.
None of those should be treated as production-ready answers for major capital delivery programmes. They are indications that the next wave of AI tooling will be embedded closer to where project information is created, checked and acted upon.
The keynote was really about context
Google said the Gemini app now has more than 900 million users, while reports from I/O described Gemini 3.5 Flash as a frontier model with stronger agentic and multimodal performance and four times faster output token speed than other frontier models in Google’s own positioning. Flash is being pushed into the Gemini app, Search, Antigravity 2.0 and the Gemini API. That matters because speed is not only a consumer convenience. In project delivery, latency affects whether a tool is used during an active coordination meeting, on a muddy site walk or inside an urgent RFI response.
The new Android XR glasses are the most physical expression of this strategy. Google’s official blog describes intelligent eyewear that can provide spoken help through audio glasses and visual help through display glasses. Audio glasses are expected first, with Samsung, Gentle Monster and Warby Parker named as partners. Google’s examples include navigation, message summaries, calls, texts, image capture, image editing with Nano Banana and translation of speech or text.
This development is useful for construction because it captures the adoption barrier. Site technology often fails when it asks people to stop the work in order to record the work. If eyewear can become light, durable and socially acceptable enough, the next field interface may be ambient. The site engineer may ask for the latest drawing revision while looking at an installation. The PM may dictate a variation note while inspecting a snag. The safety manager may capture images and request an immediate comparison with a method statement.
Why Gemini Spark deserves more attention than the glasses
The glasses are visual. Spark is operational. TechCrunch reported that Sundar Pichai described Gemini Spark as a personal AI agent that runs on Google Cloud virtual machines and acts under user direction.
“It’s your personal AI agent that helps you navigate your digital life, taking action on your behalf and under your direction,” Sundar Pichai told reporters during a product briefing.
That wording should interest project teams because the most valuable AI use cases in AEC are often admin-heavy rather than glamorous. A project manager does not need a poem about a programme risk. They need a credible draft of the client update, a cross-check against meeting actions, a reminder that an unanswered query is blocking procurement and a concise summary of the information needed before a commercial decision can be made.
Josh Woodward, vice president of the Gemini App and AI Studio, gave a revealing example in the same TechCrunch report:
“Need to send an email to your boss with a status update? Spark can pull all the facts from your emails, your docs, your sheets, and slides and write the draft for you.”
Replace “boss” with project director, client lead, design manager or QS partner. The relevance becomes obvious. AEC work is full of fragmented evidence: drawings, minutes, RFIs, cost reports, photos, risk registers, contract clauses and programme comments. A credible agent does not simply answer questions. It assembles the record, makes the next action easier and leaves a trail that a human can review.
The I/O package is better read as a set of workflow signals rather than a set of isolated product launches:
• Gemini 3.5 Flash points to lower-latency AI in live work settings. Faster responses matter when a user is inside a coordination meeting, walking a site or preparing an urgent client update.
• Gemini Spark points to persistent admin agents. The opportunity has moved beyond merely drafting text. It is pulling facts from email, documents, sheets and slides before a human approves the result.
• Android XR glasses lead to hands-free field interfaces. The caveat is practical rather than conceptual: battery life, PPE compatibility, safety policy and privacy will decide whether the tool belongs on real projects.
• Antigravity 2.0 moves towards a builder layer for internal digital teams. It suggests that construction organisations may be able to create narrow agents around their own procedures, provided those agents are governed tightly.
• Gemini Omni paves the path to a richer visual communication. Fly-throughs, image edits and video explanations could become easier, but factual control and review remain essential.
Antigravity 2.0 points to a builder layer
Antigravity 2.0 is easy to overlook because it sounds like a developer announcement. It may matter more than another chatbot feature. TechCrunch reported that Google’s agent-building tool now has an updated desktop app and command line interface and that it connects to Gemini 3.5 Flash. For construction organisations with internal digital teams, this is a clue about the market direction. AI platforms are trying to make it easier to build agents that live inside workflows rather than sit outside them.
This is where AEC leaders need to get specific. The valuable agent is not “a project management AI”. It is a narrowly scoped tool that can do one useful job reliably. Examples include reviewing whether a meeting action has the evidence required for closure, checking whether a design package has the standard documents before issue, or preparing a draft response pack for a repetitive client query. The more general the promise, the harder it becomes to trust.
Google moves too slowly, but the real problem is elsewhere
The risk for AEC is that project organisations buy the interface before they fix the information environment. Agents that have access to messy email, outdated drawings and uncontrolled spreadsheets will move faster than existing processes, but that does not mean they will move correctly. The same applies to XR glasses. Capturing more site context is useful only if teams know where that context is stored, who can access it and how it affects contractual records.
A sensible near-term adoption strategy has three parts. First, identify workflows where the decision remains human but the preparation is painful. Second, create a data access model that respects project boundaries, client confidentiality and contract requirements. Third, test agent outputs against live project examples, not vendor demos.
Google’s announcements make one point hard to ignore. The AI interface is becoming less like a website and more like an operating layer. It will sit in email, search, documents, coding tools, glass and mobile surfaces. For project delivery, that means AI adoption will become less about choosing one tool and more about designing safe patterns for AI-assisted work.
Takeaway
• Treat Google I/O as a workflow signal, not a gadget launch. The combined direction of Gemini, Spark, Antigravity and XR points towards AI embedded inside everyday project tools.
• Start with admin and evidence gathering. The strongest early use cases are likely to be status updates, meeting actions, RFI preparation, drawing search and project record assembly.
• Do not let agents inherit poor information hygiene. Data quality, permissions and version control will matter more as agents gain the ability to act.
• Use XR cautiously but creatively. Smart glasses could change field capture, but safety, privacy and durability must be tested before any serious rollout.
• Build small agents before buying broad promises. Narrow tasks with review checkpoints are easier to govern and easier to measure.
Call-to-Action
Want more practical analysis on how AI is changing project delivery? Subscribe to the Project Flux newsletter for weekly breakdowns that turn AI news into useful signals for AEC leaders, PMs, QS teams and construction technologists.
Links and Stuff
All content reflects our personal views and is not intended as professional advice or to represent any organisation.
/
1