Your AI Agent Is Useless Without This

01 /The Concept

Anon is building "the integration platform for the AI internet."

Catchy phrase, but not immediately clear. Better to see how it works through a simple example.

Say someone asks their AI assistant to remind a colleague about tomorrow's event via a direct message on Instagram. To send that message, the assistant needs to access the user's account. But how? Should the user hand over their login and password? That's obviously a bad idea from a security standpoint.

If the AI assistant uses Anon's platform, it can connect to Instagram on the user's behalf without ever receiving their credentials. When the assistant recognizes it needs to access Instagram, it surfaces an Anon-powered connection prompt – where the user enters their own login and password directly.

Once the user authenticates, a session is established between the AI assistant and Instagram. The assistant can now send the message, receive a reply, and relay the content back into the assistant interface.

That connection persists so the user doesn't have to re-authenticate every time the assistant needs to do something on Instagram. And the user can revoke it at any time – effectively through Anon's settings embedded inside the assistant.

The same flow applies to any service the assistant needs to interact with to complete the tasks it's been given.

The key technical insight is that nothing needs to be modified inside Instagram or any other external service. Anon only needs to be embedded in the AI assistant itself.

This is possible because Anon develops integration modules for popular web services. And if a specific AI assistant needs to connect to a service not yet in Anon's catalog, the developer can build that module using Anon's function library and embed it directly.

The platform currently has integrations with 10 external services, one messaging library, and is in active pilot deployment with its first clients. Despite this early stage, Anon has already closed two funding rounds totaling $6.5M – $4.5M just now, plus $2M raised earlier and disclosed simultaneously.

02 /Why It Matters

Plaid is a useful point of comparison. Plaid built integration modules for virtually every bank and financial service in the US, which third-party developers use to connect their apps to users' banking, investment, and payment accounts. Developers pay Plaid for access to that library.

In 2020, Visa tried to acquire Plaid for $5.3B – twice its valuation at the time – but the deal was blocked by antitrust regulators. A year later, Plaid raised fresh capital at a $13.4B valuation.

Anon looks like it's attempting the same story – but in the AI layer of the internet. The founder's own comparison is to OAuth-style login libraries ("sign in with Google"), but the Plaid analogy feels more apt given the structural similarity – and the valuation history worth aspiring to.

This also brings to mind AgentsForce, [covered previously](/review/tri-svojstva-dlja-bolshogo-i-denezhnogo-rynka) when it was still called Octo. AgentsForce went through Y Combinator with an AI bot for customer support teams – but it claims to resolve significantly more customer issues than comparable bots. Not because its AI model is smarter, but because its bot can independently initiate contact with other people and perform actions inside enterprise platforms rather than just answering the questions put to it.

For example: when a customer asks "where's my order?", the AgentsForce bot can look up which carrier has the shipment, contact someone (or another bot) at the carrier to find out what went wrong and when it'll be resolved, and relay that back to the customer. If the customer is still unhappy, the bot can cancel the delivery, update the order status in the database, and initiate a refund – all without a human in the loop.

The broader point: to resolve real-world problems, an AI bot can't exist in isolation. It needs to actively engage with the surrounding environment – other people, other bots, other services.

This is true for any AI assistant, not just support bots. A good assistant should be able to carry out tasks on the user's behalf in any service the user has authorized – with different permission levels for different actions (read only vs. read and write, for instance).

None of that is possible unless AI assistants can securely connect to external services with the appropriate permissions. Anon makes that connection possible. Which is why Anon is, in a real sense, a prerequisite for AI assistants becoming genuinely useful.

03 /Opportunities

Anon is early enough that there's still room to compete for the Plaid-of-AI-internet position. That alone is a valid direction.

Zoom out slightly, and the value of building protocols and platforms for AI-to-AI communication becomes clear. Soon enough, every person will have an AI assistant and every service will have an AI bot – and they'll need to negotiate with each other.

A rough sketch of what that looks like: traditional marketplaces may start becoming arenas for AI-bot interaction. A user's AI assistant finds relevant products, queries the merchant's AI bot for specs and availability, haggles on price, presents the user with ready-to-buy options, and then manages delivery and any issues that arise – all via bot-to-bot communication.

Building Anon-style integration modules for every individual merchant on a marketplace would be a nightmare. Far better to define a standard protocol that any merchant's bot must comply with – covering required functionality and message format – and pair that with a builder tool that lets merchants spin up compliant bots easily.

For product marketplaces, this is a nice-to-have. For services marketplaces – where everything constantly needs to be confirmed and clarified – it could be transformative.

Services marketplaces are a compelling category in their own right. The market is massive but poorly digitized. Today's platforms are mostly directories of business cards, where every transaction still requires painful manual back-and-forth.

Still, startups are circling the opportunity. A French startup, Scnd, [was covered previously](/review/zajti-i-vzletet) as an early mover. Lottie ([covered here](/review/a-na-rynke-uslug-jetot-shans-eshhjo-est)) and Carma are approaching the same space with interesting angles.

It's possible that the right services marketplace won't follow the traditional product-catalog model at all – it may launch as a network of communicating AI bots from day one. And that could be exactly the forcing function that finally sends this category into orbit.

Another compelling direction to run toward today, if the goal is to build something that reaches a billion in value.