Markets were flattish this week, with the S&P 500/Nasdaq 100 +0%+0%. Strong performance early in the week was driven primarily by the extension of the Iran ceasefire and hopes that de-escalation would reduce pressure on the oil complex. However, markets sold off Thursday on threats from the White House, which reignited fears of an escalating war. Crude has remained highly volatile, with WTI currently at $97/bbl, +70% YTD. Our NPM Private Market Tracker*, which shows the average price performance of the 50 largest names in our internal Tape D® data, is +12% YTD vs. SP500/Nasdaq 100 +4%/+6%. (Bloomberg; NPM)

The Rise of AI Agents

With this weekly we are touching on a topic that has been simmering in the background but that may soon move to full boil. In our view, one of the major developments in 2026 has been the emergence of “Agentic AI” (i.e. autonomous AI agents) as potentially the next wave in enterprise software. Frameworks like OpenClaw, LangChain, CrewAI, and AutoGPT are enabling or fueling anticipation of autonomous systems capable of executing real world tasks without continuous human oversight. As we will get to later, the addressable market growing rapidly: multiple independent research firms project the global AI agents market to expand from ~$8 billion in 2025 to >$50 billion by 2030, (BCC Research, 2026).

So what does an “AI agent” do? The uses are myriad. Agents, in our view, represent potentially the next phase in the evolution of AI. Imagine getting an email about setting up a meeting. An AI agent can check calendars for availability, reply to the sender with times, book the meeting, create preparation notes and send reminders ahead of time. AI agents are also beginning to transform digital commerce by acting as autonomous assistants that can search, compare, negotiate, and complete purchases. Another easy application is customer support. A customer calls about a missing package and the agent can track the package, issue a refund (if needed) and send a confirmation email. (NPM)

The OpenClaw “Moment”

Many blame OpenClaw for shining the spotlight on the AI Agent space in early 2026. OpenClaw is a development tool created by Austrian programmer Peter Steinberger in Nov 2025 as an open source autonomous agent. Despite no venture funding or marketing, OpenClaw spread quickly on developer collaboration website Github, becoming one of the fastest growing “repositories” ever. OpenClaw was notable for not requiring a high level of technical knowledge and for having easy integrations with messengers like WhatsApp, Discord and Telegram. OpenClaw had a breakout moment in Jan 2026 when perception of it tipped from a “development tool” to something that might fundamentally change how software works.

At a conceptual level, OpenClaw represents the clearest early “agentic AI”: system that can execute multi-step workflows autonomously. In essence, OpenClaw functions as a coordination layer between LLMs and external apps (APIs, browsers, email, etc.), enabling operations such as drafting communications, building playlists, managing calendars and executing follow-up actions without continuous human intervention. For instance, users have configured OpenClaw to read incoming emails and prepare draft responses to be sent either automatically or once the user signs off.

To us, OpenClaw’s significance — and why we are writing about it – lies primarily in what it signals about the direction AI, which we see as a migration from chatbots to agents. To that end, Mr. Steinberger was hired by OpenAI in Feb 2026 to help lead the company’s agentic AI efforts. OpenAI will continue to support OpenClaw through a foundation. (Forbes)

Competitive Landscape

While OpenClaw is the tool that perhaps brought agentic AI to the mainstream, there are several VC backed competitors in the space, including LangChain, CrewAI and AutoGPT.

LangChain, a private company, is an independent agent framework and has raised significant venture funding. Founded in 2022 by Harrison Chase, the company has raised $260mm in total funding across four rounds. Its most recent round, a $125 million Series B in Oct 2025, was led by IVP at a $1.25bn valuation. Prior investors include Sequoia, Benchmark, CapitalG, Sapphire Ventures, Workday, ServiceNow, Cisco and Databricks. (Pitchbook)

In addition, large technology incumbents are actively integrating agentic capabilities into their ecosystems.

As mentioned OpenAI, effectively “acqui-hired” OpenClaw and has made AI agents a strategic priority. OpenAI’s long-term roadmap explicitly frames agents as the next phase of its product platform — moving from subscriptions toward usage-based revenue tied to autonomous workflows. (Bloomberg; OpenAI)

Microsoft, via Copilot Studio, is integrating agent capabilities into its productivity suite. Microsoft’s advantage is distribution: embedding agents into tools already used by hundreds of millions of workers creates adoption pathways unavailable to independent frameworks. (Microsoft; Bloomberg)

Google is investing heavily in agents through its Cloud Agent Builder and Gemini model integrations, particularly targeting enterprise customers on Google Cloud. Another example is Salesforce’s Agentforce platform, which provides autonomous agents for sales, service, marketing, and commerce. In addition, Amazon Web Services offers Bedrock Agents as part of its cloud AI stack.

Gartner has predicted that 40% of enterprise applications will be integrated with task-specific AI agents by end of 2026, up from less than 5% in 2025,. (Gartner)

Market Size

Analyst estimates for the AI agents market converge around a consistent picture of rapid growth, even as precise figures vary. Markets and Markets estimated the market at $7.8bn billion in 2025, projecting growth to $53bn by 2030 at a CAGR of 46%. Grand View Research projects the market reaching $50bn by 2030 and BCC Research projects the market growing from $8bn n 2025 to $48bn by 2030.

Risks and Limitations

While we believe AI agents are headed towards the mainstream, we see several key risks. First, technical limitations remain. While agents excel at certain workflows, complicated tasks remain challenging. Currently, many deployments still require meaningful human oversight, such as proofreading an agent generated email response before it is sent.

Security is also a central concern, as agents often require access to personal information — email, financial accounts, databases — creating potential vulnerabilities.

Thirdly, reliability is also a significant challenge. Even a small error rate could have significant consequences. A simplistic example is an AI shopping agent that purchases 100 pairs of shoes instead of just one. Another analogy is the handful of autonomous vehicle accidents that have hindered the development of that space for years.

The table below shows estimated performance of various private secondary market sectors YTD based on NPM’s proprietary Tape D data. The leader YTD is industrials, which includes such “hot” sectors as space, robotics and defense.