Agents Go In-App. Infrastructure Scales. Identity Takes Over.

What happened
Microsoft said Microsoft 365 Copilot agents can now surface business apps directly inside Copilot chat so users can move from insight to in-app execution without leaving the conversation.

Why it matters
This is the clearest signal yet that Copilot’s endgame is workflow orchestration, an Agent Store for distribution plus MCP Apps / Apps SDK for building custom agent integrations and admin-center controls to govern it.

What’s next
Expect a burst of MCP-based integrations from ISVs and internal IT teams. The winners will be the ones that make permissions, supervision, and auditability default, not optional.

What happened
Cloudflare announced an expansion of its Agent Cloud aimed at running large numbers of long-running agents, including Dynamic Workers (isolate-based execution), Artifacts (Git-compatible storage), Sandboxes (persistent Linux environments), and Think (a persistence-focused Agents SDK framework).

Why it matters
The agent bottleneck is no longer model capability - it’s durable, affordable, secure execution (compute + storage + an OS when needed) that can scale when every user has “dozens of agents” running concurrently.

What’s next
Watch how quickly developers adopt Dynamic Workers/Sandboxes and whether Cloudflare’s “single pane of glass” model catalog with easy provider switching actually becomes the default neutral layer for agent workloads.

What happened
SDxCentral reports Cisco is looking to acquire Astrix Security for up to $350M. Astrix focuses on identifying and securing non-human identities (including AI agents) and even inventories MCP servers, adding permission context and automated risk scoring to prioritize remediation.

Why it matters
As agents gain tool access, “identity” becomes the control point: the attack surface is no longer just humans and APIs, but fleets of semi-autonomous actors with credentials, scopes, and shadow deployments.

What’s next
If this closes, expect a fast-follow consolidation wave in agent security (identity, observability, and policy enforcement) as vendors race to become the gatekeeper for non-human privileges.

What happened
Stanford HAI’s AI Index takeaways highlight several inflection points: generative AI reached 53% population adoption within three years (with the U.S. ranking 24th at 28.3%), the Foundation Model Transparency Index average dropped to 40 from 58, and U.S. trust in government AI regulation measured 31%.

Why it matters
The center of gravity is moving from “can we deploy AI?” to “can we govern it and absorb the labor + infrastructure shock,” with the report also pointing to entry-level pressure (ex. 22–25-year-old software developer employment down nearly 20% since 2024).

What’s next
Expect enterprise spend to re-balance toward defensibility, provenance, audit trails, restoration, and safety controls because adoption is already mainstream while transparency is moving the wrong way.

What happened
Commvault announced new and forthcoming Commvault Cloud capabilities—Data Activate, AI Protect, and AI Studio—positioned to help enterprises activate AI safely, govern AI agents, and build agentic workflows with trusted data, governance, and recovery.

Why it matters
This is an enterprise pattern you’ll see everywhere in 2026: “AI resilience” becomes a first-class requirement, meaning teams need to not only protect data but also curate AI-ready datasets (Iceberg/Parquet) and roll back agent-driven changes across systems.

What’s next
Look for Commvault and peers to compete on two things: agent discovery/governance coverage and “full-stack recovery” playbooks that treat agent configs + dependencies as recoverable assets.

What happened
Corvus Robotics announced Corvus Trident, an AI-powered device mounted on forklifts and other material handling equipment to automatically capture pallet movement, reading multiple barcodes at once and tracking pallets/equipment in real time without manual stop-and-scan steps.

Why it matters
Physical AI fails when the system of record diverges from the real world. Trident targets the missing scan problem that drives inventory drift, shipment errors, and costly disputes.

What’s next
Expect more instrumentation first autonomy in warehouses - better ground truth now, more automation later, especially where legacy WMS data is inconsistent.

What happened
Circus SE announced it fully placed its first asset-backed bond via FINEXITY. The bond is backed by Circus’ autonomous meal-production AI robots, and the company signed an additional agreement to pursue further financings up to €50M to support rollout for defense and civilian customers.

Why it matters
Real-world robotics scales on capital, not just code. Asset-backed financing is a signal that the market is starting to treat deployed robots as financeable equipment fleets (predictable output, collateral, structured vehicles).

What’s next
Watch whether this model spreads to other robotics categories (warehouse, field service, security) as they push from pilots into repeatable, financed deployments.

Agents are no longer interfaces—they’re operators inside your systems.
The real battle is shifting to who controls their access, execution, and recovery.

Pick one workflow you do daily (CRM update, ticket triage, reporting).

Recreate it using an agent inside a tool (You.com, Copilot, ChatGPT, or Claude)
Give it permission to act, not just respond.
Then ask: what breaks if this runs 100x per day?

You’ll quickly see the gap between “AI demo” and “AI system.”