The Last Mile of Agentic Commerce: Why Payments, Privacy, and Trust Will Define Enterprise AI Adoption

Agentic commerce is not here yet, with Google Payments or Link still inaccessible from MCPs or easy OAuth SDKs for agents at large.

But it’s getting there. Shopping is already scaling – and with DTC, the layered tiers will bubble up to B2B and the mid-market soon enough.

As Mercury switched to Column recently, I was left with 30+ subscriptions across 2 companies to move over in about 2 weeks. July was more intense than ever, with Amazon, Google Workspace, Meta, Lovable and Replit, OpenAI relied on credit cards auto migrated to the new bank account.

Our agentic system couldn’t support the journey since most Anthropic Cowork agents still need 2FA or Google Sign In protection for sensitive billing details. That’s only natural.

Our Claude SDK and NanoClaw agents can’t access Google Payments/Links despite the progress made by Google and Stripe in this direction – or having access to Apify and the newly announced agentic payments via x402.

Similarly though, downloading Meta invoices via an API is just impossible. Finance requires manual transactions pulled from the business manager manually. It is what it is.

Some of these old-school manual processes are productivity and efficiency killers. It’s also a liability offloading to a handful of people. C-level staff members aren’t supposed to spend time on that. And FTEs with access to sensitive data may backfire – even 1 in 100 is enough to cause epic damage in that department.

The thing with sensitive data and details is – it’s “usually” not a problem, until you hear a story of $80K wired to a personal account randomly or a flagged bank account, or 2 years back of transactions getting lost to random firms month after month. Even though these are rare, this can break too many businesses.

At a higher scale, access to customer lists or bank accounts or price rates can be the ultimate competitive advantage to land a job with a competitor.

Agents that can “guarantee” privacy solve far more than an individual can imagine. We suspect OpenAI and Anthropic would be running smaller models offline or have hosted versions for enterprises and even SMEs that are fully controlled and safe for situations like these.

In hindsight – agents are here to stay. Efficiency is going up for the PE companies we work with. Some cases are still unresolved (like billing) while others already deliver 100x efficiency easily, replicated on multiple fronts.

And with 20,000+ PE firms in the US alone, we’re seeing this playing out in a new portfolio company every 2 to 3 months for the past 2.5 years, and counting.