OK, Let's Talk About OpenClaw

I round up the most relevant AI-in-finance news - the deals being done, who’s rolling out what, and what’s actually working on the front lines

…and compared OpenClaw to Linux. Nvidia launched NemoClaw, their enterprise-grade version, the same day. "Claws" have officially gone mainstream.

JPMorgan halted a $5.3 billion debt deal for Qualtrics after investors balked at AI disruption risk. HSBC is weighing 20,000 job cuts tied to AI automation. PwC's US boss told the FT that partners who resist AI "won't be here that long."

Elsewhere: Bridgewater's chief scientist defected to DeepMind, Anthropic is now winning the majority of new enterprise AI spending, and "vibe coding" made Collins Dictionary's Word of the Year.

But first, my take on what a "claw" is, how it fits into the AI landscape, and what it means for finance.

In This Week’s Issue:

From The Trenches

What Is OpenClaw?

OpenClaw has been around since November 2025. It started circulating in small developer circles on X around Christmas before properly blowing up a couple of months ago. You've probably seen some of the bizarre stories. Agents creating their own Reddit channels. In China, retirees are using it to raise lobsters.

Underneath the novelty, though, is a technology with real practical use cases. And when the CEO of Nvidia is on stage at GTC announcing their own enterprise-grade version (NemoClaw), it's probably something worth paying attention to. Anecdotally, it's started coming up on a couple of calls I've had over the last few weeks. OpenClaw has clearly hit the mainstream.

Whilst the parts of what OpenClaw does have been around for a while, the ease and distribution of it are what's taken it to the next level. Here's the progression.

Step 1: Chat. ChatGPT, Claude, Gemini. You type a question, you get an answer. Incredibly useful, but fundamentally passive. The AI talks. It doesn't do.

Step 2: Chat with tools. Perplexity, ChatGPT's plugins. Same conversational interface, but now the AI can reach out and touch things. Browse the web. Read your files. Pull data from connected apps. But you're still sitting in the chat window, driving the conversation.

Step 3: Chat with connected apps. Claude Cowork is the best example here. Same conversational interface, but now it connects directly to your work tools: Google Drive, Slack, Notion, your CRM. It can read, write, and take actions across them. A lot of people I'm speaking to are already using Cowork. It's a meaningful jump, but you're still initiating every task.

Step 4: Agents. Claude Code, Devin, and similar tools. The AI doesn't just use tools when you ask. It plans, executes, and iterates autonomously. Give it a task, it figures out the steps, does them, checks its work, and comes back with the result. These tools are typically built for developers. Command-line interfaces. Configuration files. API keys.

Step 5: Accessible agents. This is OpenClaw. Same underlying concept as Step 4. Same AI models. Same tool-calling capability. But accessed through WhatsApp, Telegram, Signal, or Discord. It's a contact in your messaging app. You message it like you'd message a colleague. "Find me flights under $500 to New York next week and book the cheapest one." It goes and does it.

OpenAI acquired OpenClaw's creator in February and moved the project to an open-source foundation. Anthropic launched features like remote control for Claude Code and Dispatch for Cowork that mirror many of the same capabilities. Send Claude a task from your phone, it executes on your desktop. Same idea, different packaging.

Then Jensen Huang used GTC to declare that every company needs an OpenClaw strategy and launched NemoClaw to make it enterprise-ready. Tencent integrated it into WeChat for over a billion users. Alibaba launched Wukong. Baidu built agents on top of it.

When Anthropic, Nvidia, OpenAI, and Tencent all converge on the same idea in the same month, it's not a trend. It's the new default.

The WSJ catalogued what OpenClaw agents are doing right now: morning briefings, email triage, meeting summaries. Useful, but generic. The finance use cases are where it starts to get more interesting.

When anyone asks me for a use case, my basic reply is: "Well, what are you asking your associates for at the moment?" Because that's what this is starting to look like. An associate or analyst that's available to you whenever you need them. The range runs from simple, one-off requests to more complex, multi-step workflows. Quick access to deal information mid-meeting. "What was revenue in 2025 and how did that compare to 2024?" Real-time answers pulled from your deal platform, not from someone scrambling through notes. Scheduled monitoring: checking for portfolio company news, flagging management changes at competitors, surfacing regulatory filings, and sending you a prioritised briefing before you've finished your coffee. And then more complex tasks: sourcing new opportunities with specific criteria, cross-referencing public data against your deal database, keeping the results updated in a structured file or system.

That list of things you can delegate, from one-off requests to structured multi-step work, is only going to compress over the next few months. What takes an hour today will take minutes. What takes a day will take an hour.

What this highlights, though, is how important it becomes for these agents to connect to platforms that hold relevant context and can store and log everything that's going on. As you have more agents working for you, you need to know what they're doing. You need that work logged, structured, and visible so that multiple people on your team can see it, collaborate on it, and stay on the same page. This continues to be a core tenet of what we're building at DealSage: the context-connected, infrastructure that agents plug into, pull from, and write back to reliably.

It's not free. OpenClaw is open source, but the AI models it connects to are not. Every query costs money via API calls, and unlike a $20/month subscription, costs scale with usage. For a firm deploying this across a team, costs can grow fast.

It's painful to set up. Hours of configuration. Connecting tools, setting up authentication, handling edge cases, maintaining it when things break. Not plug-and-play yet.

And for anyone in finance: security. You're giving an AI agent access to deal data, financials, client information. The open-source version has no enterprise security model. That's exactly why Nvidia built NemoClaw. But even the enterprise versions are early.

What is clear, though, is that the access layer is arriving. The gap between using ChatGPT for summaries and having an agent pull deal data mid-meeting is closing fast. Ultimately, the agent is only as good as the infrastructure it connects to and coordinates with but we're well on our way.

News Digest

JPMorgan Halts $5.3bn Qualtrics Debt Deal as AI Fears Chill Demand

Banks led by JPMorgan paused a $5.3 billion debt package for Qualtrics this week after failing to find enough investors in the leveraged loan and junk bond markets. The deal was meant to fund Qualtrics' $6.75 billion acquisition of healthcare data firm Press Ganey Forsta. Instead, it's become the most visible example yet of AI disruption fears directly killing a deal.

One junk bond trader told the FT: "Software is a tough sell right now. It wouldn't be a deal that we want to participate in anyway."

The details:

Why it matters: This isn't a startup struggling to raise. This is a $12.5 billion PE-backed company with a consortium of the world's biggest banks behind it, and they still couldn't get the debt away. AI disruption risk is now priced into credit markets, not just equity.

My take: The hung deal risk is the story. Eleven banks committed bridge financing. If they can't relaunch, they're holding $5.3 billion on their own balance sheets. But the deeper signal is what this means not just for software, but for any business in a sector that AI could disrupt. This is the first time I've seen AI fear translate into a real impact on a large, established company beyond stock market volatility. When bond markets start reacting like this, it's a different level of conviction. As I argued two weeks ago, this is only the beginning.

Read more

HSBC Mulls 20,000 Job Cuts in AI-Fuelled Overhaul

HSBC is weighing deep job cuts that could affect around 20,000 roles, roughly 10% of its global workforce, over the next three to five years. CEO Georges Elhedery is betting on AI to shrink the bank's middle and back offices as part of a broader restructuring announced since he took the role.

The details:

Why it matters: This is one of the largest AI-driven workforce restructuring plans announced by a major bank. When a bank with 208,000 employees says AI could eliminate 10% of roles, the ripple effects extend to every financial services firm asking the same question internally.

My take: The "three to five years" framing is doing a lot of heavy lifting. It gives HSBC cover to start immediately while sounding gradual. The real tell is the focus on middle and back office: processing, compliance checking, data entry, reconciliation. These are the roles most firms have been quietly automating already. HSBC is just being more explicit about the trajectory than most.

Read more

PwC US Boss: Partners Who Resist AI "Won't Be Here That Long

PwC's US chief executive Paul Griggs told the FT that partners who aren't "paranoid about being AI-first" will probably be replaced. But the headline isn't the interesting part. The interesting part is what PwC is actually doing about it.

Griggs announced PwC One, an AI platform offering automated services to clients, including M&A due diligence and tax tools. The pricing model is shifting from billing by the hour to subscription or consumption-based pricing. Some of these services will be available "without a PwC person in the loop." A Big Four firm, building AI tools that clients access directly on a subscription, without a consultant involved. That's a fundamental shift in the delivery model.

The details:

Why it matters: The Big Four billing model hasn't fundamentally changed in decades: thousands of junior staff doing relatively routine work billed at $300+ per hour. PwC is openly acknowledging that much of that work can be automated and sold as software. If the delivery model changes at PwC, it changes everywhere.

My take: The shift from hourly billing to subscription-based AI tools is a preview of where professional services is heading more broadly. When you can package expertise as software and sell it without a human in the loop, the economics of the entire industry flip. Margins expand for the firms that move first. Fees compress for everyone else. The FT comments section captured the tension perfectly: "People who don't embrace AI have no place at the firm, people who do embrace AI will have no place at the firm." Nobody's resolved that yet. But PwC is at least being honest about the direction of travel.

Read more

Other Interesting Things I’ve Read of Seen This Week:

Bridgewater's chief scientist Sekhon joins Google's DeepMind as Chief Strategy Officer (Mar 18) - Jasjeet Sekhon, head of AI at Bridgewater Associates, is leaving for DeepMind. He said he believes DeepMind is "best positioned to develop AGI safely." When the world's largest hedge fund loses its chief scientist to a frontier lab, the talent war has a clear winner.

Jensen Huang wants every company to have an OpenClaw plan (Mar 18) - "Every company in the world today needs to have an OpenClaw strategy. This is the new computer." When the CEO of the world's most valuable company starts comparing an open-source project to Linux, the VCs who funded proprietary agent startups get very quiet.

OpenAI plans desktop 'superapp' to simplify user experience (Mar 19) - OpenAI is merging ChatGPT, Codex, and its Atlas browser into one desktop app. Chief of Applications Fidji Simo admitted they were "spreading our efforts across too many apps and stacks." Even OpenAI is trying to solve the accessibility problem.

The rise of 'vibe coding': Claude Code, Cursor, and Codex battle for developers (Mar 20) - "Vibe coding," where developers lean on AI to generate and ship code from prompts rather than writing it line by line, was named Collins Dictionary's Word of the Year. If you want to understand Step 3 of the ladder I described above, start here.

OpenClaw enthusiasm grips China as schoolkids and retirees alike embrace AI agents (Mar 19) - Grandparents raising lobsters. Schoolkids automating homework. Tencent integrating it into WeChat for a billion users. China is moving faster on agent distribution than anyone in the West. And they're having more fun doing it.

Edward Jones insists AI will not replace its $2.5tn financial adviser network (Mar 20) - CEO Penny Pennington told the FT that "a machine can make a pretty darn good investment recommendation" but "it doesn't replace the financial adviser." The other side of the PwC coin. One firm says adapt or leave. The other says humans are irreplaceable. They can't both be right.

Acquisition Intelligence is a weekly newsletter on AI in M&A for finance professionals, private equity investors, investment bankers, corp dev teams, and deal-makers.

For questions, feedback, or to share what you're seeing in the market, reply to this email.

P.S. I'm Harry, co-founder of DealSage. We're building an AI-native deal intelligence platform to help professionals turn their institutional knowledge into better decisions. If you're curious what we're up to, check out dealsage.io or just reply here