AI Agents

Vertical AI agents displacing human workflows in regulated industries

The displacement of human labor in regulated verticals has moved from pilot to production scale. Basis raised a $100M Series B at a $1.15B valuation to automate accounting and audit workflows, while Harvey — trusted by 142,000 lawyers across 1,500+ organizations — continues expanding agentic legal workflows backed by Sequoia and a16z. Genspark raised a $100M Series B extension at a $2.6B valuation for agentic workplace software, and Poetic is automating high-stakes workflows like fraud investigations and underwriting at just one year old. a16z's new partner Joe Schmidt authored frameworks specifically mapping how AI products gain durable footholds in high-stakes enterprise verticals, signaling the firm's conviction that regulated-industry agents are the next major platform.

The picks-and-shovels layer beneath AI agents is attracting serious capital independently of any application layer. RadixArk raised $100M in seed funding to commercialize the SGLang inference engine and Miles RL training framework, with customers including Google, Microsoft, and Nvidia. Mem0 provides production-ready agent memory infrastructure, Letta (born from UC Berkeley's MemGPT project) is building stateful persistent agents, and Judgment Labs is building evaluation infrastructure for long reasoning traces. The MCP protocol has become a de facto standard, spawning dozens of new tools — from API-to-MCP converters to MCP-native analytics platforms like Amami — creating an ecosystem of interoperable agent tooling.

Why it matters · Infrastructure layers historically compress margins at the application layer while generating durable, high-margin businesses — investors who miss this layer risk overpaying for application-layer exposure.

The market has decisively shifted from single-agent demos to coordinated multi-agent systems as the production standard. Munder Difflin (143 Product Hunt votes) lets users run autonomous AI clones via Claude Code and Codex 24/7; Hoplite deploys cloud-based parallel coding agent factories; and AgentGrid provides infinite-canvas orchestration for role-based agent teams (builder, QA, reviewer, devops). Google's Gemini 3.7 Flash — launched by Sundar Pichai and collecting 181 Product Hunt votes — is explicitly optimized for 'coding and agents,' while DeepSeek Harness offers a composable open-source agent runtime with fully pluggable components, reflecting that frontier model labs are now competing on multi-agent capability as a primary axis.

Why it matters · Teams shipping multi-agent orchestration frameworks now will own the workflow abstraction layer, making it costly for enterprises to switch once agentic pipelines are embedded in operations.

As agents gain autonomous access to enterprise systems, a dedicated security and governance market is crystallizing around them. NewCore AI raised $66M at a $300M+ valuation for AI agent security; Lyrie.ai delivers autonomous cybersecurity for the AI agent era; Darkmoon deploys 18 specialized AI agents for autonomous penetration testing; and Bloom Security, Onyx Security, and Geordie AI are each building distinct layers of agent oversight. Open-source projects like HOL Guard (agent firewall), Execlave (runtime policy enforcement), and Phinq (risk-classified tool-call governance) are signaling grassroots demand ahead of enterprise budget formation.

Why it matters · Every enterprise deploying autonomous agents will need a security and governance stack — this is a mandatory spend category, not discretionary, giving early vendors structural pricing power.

A wave of free, open-source coding agent platforms is directly targeting paid incumbents. Freebuff (199 Product Hunt votes, CTO Victor Cheng's tagline: 'kill Claude, Cursor, Replit, and Devin') offers CLI, desktop, and web interfaces with no subscription costs. OpenCode brings agentic AI to local/self-hosted models with file, git, and terminal support. AgentLoop, HAR, and PenguinHarness each provide open-source multi-agent orchestration frameworks. This dynamic mirrors the historical open-source playbook: commoditize the complement, capture the ecosystem, then monetize infrastructure — putting margin pressure on Cursor, Replit, and Cognition's Devin while benefiting model providers like Anthropic and OpenAI whose APIs these tools consume.

Why it matters · Investors holding positions in closed-source coding agent incumbents face accelerating commoditization risk; the defensible value is shifting upstream to model providers and downstream to vertical workflow integrations.