AI Agents

Vertical AI agents are displacing human workflows in regulated industries

The most capital-intensive bets in AI Agents are targeting high-stakes, compliance-heavy verticals where replacing human labor creates the clearest ROI. Hippocratic AI has completed over 115 million clinical patient interactions at a $3.5B valuation and $404M raised, while Legora serves 1,000+ law firms across 50 markets with an agentic legal OS. Harvey is trusted by 142,000 lawyers at 1,500+ organizations, backed by Sequoia and Andreessen Horowitz. Basis just hit a $1.15B valuation with a $100M Series B led by Accel for AI agents targeting accounting firms. The pattern is clear: vertical agents that can credibly operate within regulatory guardrails — healthcare, legal, finance — are attracting the largest rounds and most durable customer bases. Saris automates multi-turn back-office workflows for financial institutions, while LogicFlo AI targets regulated biotech and pharma environments, signaling the vertical agent wave is far from saturated.

A distinct meta-layer of tooling — memory, evaluation, observability, security, and identity — is emerging to support the proliferation of autonomous agents. Letta (born from UC Berkeley's MemGPT) is building stateful persistent agent memory; Mem0 offers production-ready agent memory with three-line integration; Judgment Labs and BentoLabs address evaluation and monitoring of long-running agents; Keycard provides identity infrastructure for autonomous systems; and Trajectory's continual learning platform already counts Harvey, Clay, Decagon, and Mercor as customers. Agent Mail, backed by General Catalyst and Y Combinator, rebuilt email infrastructure from scratch for agentic workflows. This infrastructure layer is accelerating because agents increasingly operate autonomously across multi-step tasks, creating silent failure modes that existing DevOps tooling cannot detect.

Why it matters · Infrastructure plays in prior platform shifts (cloud, mobile) became some of the most defensible businesses; the agent infrastructure layer is at a comparably early and high-leverage moment.

Claude Code and Codex are reshaping the developer tooling landscape so rapidly that an entire ecosystem of observability, cloud execution, and orchestration tools has emerged around them. Backplanes provides session reporting for Claude Code and Codex agents; Replicas runs coding agents in isolated cloud VMs triggered from Slack, Linear, or GitHub; Limrun cloudifies local development utilities so coding agents can compose capabilities on the fly. Anthropic itself reports that over 80% of code merged into its codebase was written by Claude, though a 100,000-developer GitHub study cited by The Generalist shows productivity gains collapse from 300% file edits to just 30% more shipped software — suggesting the tooling layer to close this gap is the real prize. OpenAI's acquisition of OpenClaw, an open-source agentic coding framework, underscores how seriously model labs are treating coding agent distribution.

Why it matters · The gap between raw coding-agent productivity and actual shipped software represents a multi-billion-dollar tooling opportunity for whoever can close the 'last mile' of agentic development.

A cluster of AI agent startups targeting cybersecurity and defense has moved from concept to funded product. NODA AI is building AI-native orchestration for cross-vendor autonomous defense systems spanning air, land, sea, and cyber domains; Breaker Industries deploys onboard AI agent software for military operators commanding robot teams via voice with no cloud dependency. On the commercial security side, Artemis Global Technologies targets real-time threat detection and automated response as a SIEM replacement, while Cogent Security uses AI agents to bridge vulnerability detection and remediation. Lyrie.ai is building an autonomous cybersecurity platform for the AI agent era itself — a sign that agent-on-agent threat models are already being priced in.

Why it matters · Defense and cybersecurity represent two of the few markets where buyers will pay for reliability over cost — creating durable pricing power for agent platforms that can demonstrate mission-critical performance.

Enterprises are moving beyond single-agent deployments toward coordinated multi-agent systems, and a new class of orchestration platforms is emerging to manage them. AgentOS offers a local-first control surface for managing agents, tasks, workspaces, and approvals; Onpilot AI deploys specialized AI workers across 3,000+ integrations via Slack, Teams, and WhatsApp; Axiamatic builds an agentic control plane for enterprise transformation programs. Salesforce's Agentforce platform — already at 12,000 customers including Dell, FedEx, and PepsiCo — is setting the enterprise baseline, while Dust serves 5,000+ organizations with a multiplayer AI workspace. The $9.16B deal week of May 25 and sustained deal velocity (114 deals in 28 days) suggest enterprise buyers are actively committing budget, not just piloting.

Why it matters · The orchestration layer that becomes the default enterprise control plane for multi-agent deployments will capture recurring revenue at every agent interaction — a structurally superior position to single-agent point solutions.