Factor Capital Update - February 2026

This month had three major events worth discussing: OpenClaw, the crypto crash, and Claude Code.

If you scan X right now, the most talked-about thing online, aside from the Epstein files, is the explosive adoption of OpenClaw (formerly called Moltbot or Clawdbot). The name keeps changing, but the signal is clear.

OpenClaw is an open-source desktop application that operates with complete agency across any apps you grant it access to. When given permission to access your email, calendars, messaging apps, browser, and login credentials, the agent can autonomously handle tasks. It summarizes emails, schedules meetings, books reservations, and performs virtually any computer-based task a human could.

The early response has been remarkable. People are astounded by its capabilities because it represents what most expected Siri to become. It is a genuinely intelligent, constantly learning virtual assistant that takes natural language commands and executes complex workflows with ease.

The Difference Between Chatting and Doing

For those not deep in the technical weeds, it is critical to understand the difference between a Chatbot (like the ChatGPT interface you're used to) and an Agent.

Think of a Chatbot like a very smart consultant locked in an empty room with a typewriter. You slide a question under the door asking for a dinner reservation strategy, and the consultant types back a brilliant plan. But they cannot leave the room. They cannot pick up a phone. They cannot actually make the reservation.

An Agent is that same consultant, but you've given them the keys to your office, your unlocked phone, and your credit card. You don't ask for a strategy. You say, "Get me a table for four at 7pm." The Agent opens the app, clicks the buttons, enters the details, and sends the confirmation to your calendar. It has "hands" to manipulate the world on your behalf.

That is what OpenClaw is. It doesn’t just generate text. It executes actions.

Claude Code and the Terminal

This tool didn't spawn from nowhere. It is the downstream utility of command-line agents, most prominently Claude Code.

Claude Code launched as an app that runs in your computer's terminal, powered by the latest Anthropic models. Google Gemini and OpenAI have their versions (Gemini CLI and Codex), but the shift here is fundamental. Previously, LLMs were great coders but producing anything beyond a single script was challenging. You’d ask all the things you needed to do to get something built and launched, but you were doing the work; it was just giving you the code or editing yours. Now, through simple prompts, the agent just gets to work, and you never touch or see a line of code from that point on.

To understand the power here, you need to understand the environment. Most people interact with computers via the GUI (Graphical User Interface). This is your dashboard. It has icons, windows, and mouse clicks. It is designed to be safe and easy, which means it hides 90% of what the computer can actually do. The Terminal is the text-based interface below the graphics. It is where the computer actually runs.

Giving an AI agent access to the Terminal grants it total control. It can do everything a user does in a browser, plus everything a developer does in the system backend. It can interact with the internet and APIs while simultaneously rewriting system files or installing software. It bypasses the limitations of visual apps to control the machine and the network directly. On top of this, Claude Code runs autonomously for hours. Give it a task, and it plans the approach, performs research, and maps out dependencies. It constructs a to-do list, then spawns a set of parallel agents to tackle it like a flock of engineers.

Vibe Coding: The Velocity Shift

I have talked about vibe coding here before, so I won't rehash the basics. But we need to update our mental models because the capability has advanced significantly in just the last few months.

The models are smarter now. But the real shift is that you stop coding and start managing a fleet of developers.

The leverage here is massive. A single skilled engineer is now effectively a 10x engineer. They don't type faster. They orchestrate agents that write code simultaneously across different parts of the stack.

For the non-engineer, the floor has risen dramatically. You are now, conservatively, as skilled as an entry-level software engineer.

Exhibit A: FamilyFlix

I built FamilyFlix in an hour using Claude Code because our 4-year-old was having nightmares after watching the savage animals in Zootopia, and I wanted a tool to help us determine the "scariness" of movies when trailers were misleading. Now, we can type in any movie title, and FamilyFlix generates a personalized AI-powered rundown of whether it’s safe to screen for the kids, helping us get a full night’s sleep.

A year ago, that was a full weekend project hardly worth doing, and it would be functional but not amazing. Now, it's a 1-hour detour, and if you want to take it a step further, another day of work to publish it to the app store and generate all your marketing materials and videos automatically.

This is how quickly things are moving.

The SaaS Ripple Effect

This is the reason why SaaS stocks are down roughly 15% YTD as a group. Investors are increasingly questioning how sticky enterprise software licenses really are in a world where these types of apps are so easily accessible.

There are two threats here. First, applications can be rewritten in-house through these agentic software tools. Second, the core functions of expensive software are being made irrelevant by ever-improving models. Do you really need an Adobe license when free open source models can generate pixel-perfect edits to your photos and create professional marketing collateral with a few simple prompts?

The moat is drying up.

The Screen-Free Future

As has been the case with all forms of Gen AI since the release of ChatGPT, these open-source tools are just a preview for the mainstream future.

In the early days of ChatGPT, we had tools like AgentGPT. They were hacky open-source projects that let you string together model calls with terrible UX. Months later, those features just became standard parts of the model offerings.

We are seeing the same thing now. We are moving toward the new screen-free interface I mused about last year.

Remember, OpenClaw is open-sourced so anyone can use it to build a better, more secure, more user friendly version you pay for versus the hobbiest, hacker version or today. You will use something like “PaidClaw” on your phone to talk to a securely managed version of OpenClaw. You will chat with it, grant specific permissions, and finally get the virtual assistant experience that was promised and never delivered with Apple Intelligence.

This changes the winners and losers in the app economy.

The Winners: Systems of Record. The things that hold your actual data and are your means of interacting with the outside world become critical. Data Lakes, ERP systems, Gmail, Photos, Contacts, Maps, Calendar, Notion. These are the sources of truth your agent needs to access to answer questions for you with contextual intelligence.

The Losers: Interface Apps. Apps like OpenTable will be in an interesting spot soon. I don't need their interface if my agent can just talk to their API. Or, using services like Twilio, my agent can literally call the restaurant and speak to a human to make a reservation for me. The interface becomes invisible.

The Crypto Divergence: Crash vs. Utility

This brings us to the third event of the month: the crypto crash.

Despite the Clarity Act bringing long sought regulatory clarity to crypto projects and token issuers, tokens are getting destroyed right now because liquidity has been sucked from the market. The speculators have moved on, and the crypto casino has lost its monopoly on trader attention. But while the price action is ugly, the actual utility for crypto has never been clearer. It is just... boring.

The dominant use cases emerging are payments and tokenization. Specifically, agents need money. But the reality is, you probably don’t want to give an autonomous bot your bank login.

If an agent has agency, it can make mistakes. It can make poor decisions based on assumptions or lack of information. You do not want a rogue robot having unlimited access to your credit line. Crypto solves this. You can spin up a wallet for an agent and fund it with $500 USDC. The agent can spend it programmatically to book flights or pay for API calls, but it cannot spend a penny more than what is in the wallet.

Our portfolio company Nevermined is working on exactly this by partnering with companies like Visa and Cloudflare on agent-to-agent payment orchestration. It is a massive, critical infrastructure play. But it doesn't sell a "narrative" or pump the next cycle like decentralized social networks or ZK identity did in the past.

The Builder's Paradise

In total, as I said at the beginning of the year, this is an amazing time to be a user of technology.

The possibilities are thrilling. Best-in-class companies are shipping faster than ever because they are using Claude Code to automate their own development. They are churning out features at a record pace. Users win as we keep getting better products for the same price.

But it remains to be seen how this shakes out for the workforce. Some companies will view this as an opportunity to accelerate their existing teams. Others will find opportunities to cut large components of a workforce that can be automated.

If you have a fully functional consultant sitting in your terminal for a trivial expense, the math changes. Business owners will look at their employees and ask hard questions. Which people are just doing repetitive tasks? Filling out forms. Submitting claims. Scheduling. These workflows can be taught to an AI assistant in minutes. You start to question which employees are actually required.

We also do not know which applications are immune to this disruption. I mentioned OpenTable above. It is possible that these become interface-free agent-to-agent coordination layers. When I need a table at 7pm, my agent talks to the OpenTable agentic server coordinates the reservation.

Or maybe the aggregation model breaks entirely. Restaurants might plug in an open-source booking agent that’s equally discoverable and stop paying OpenTable fees altogether. Agents do not care about a sleek centralized UX. They just research reviews and use my learned preferences pick the spot and then ping the restaurant's agent to coordinate the booking directly. This is already occurring with software development, where Claude will prescribe the tech stack I should use, install the packages on my computer to communicate directly with those services, and then coordinate everything on its own - sales and marketing roles evaporate when the model is just picking the best solutions through unbiased analysis of their capabilities and fit.

For investors, the landscape is shifting so rapidly that predicting how far your typical consumer and business operator will go with adopting these innovations is challenging. However, the trends we observe at the edge today are likely to reach the mainstream in the coming months. Operators of “traditional” businesses have a significant opportunity to become the downstream beneficiaries of this activity if you leverage the advantages and deflationary effects they provide.

Thanks as always for reading,

Jake

Jake Dwyer
Founder
Factor Capital