This is where you get EXCITED about agentic payments

Consumer internet is all about consumption - consumption of information, and services. Needless to say, AI has completely taken over consumption of information in the past couple of years. 

We are at the cusp of AI taking over consumption of services via agents. It is not trivial since services need payment. Recent advancements in AI have been all about unstructured generation, and payment is a little far from that archetype. 

Not only do we need extreme reliability, we need to rethink financial infrastructure ground up. Here are examples of some fundamental problems that we run into while imagining agentic payments with existing infrastructure

1. Would you give your agent access to your bank account?

2. It’s okay for banks to block your account if it sees five $2 transactions in 10 seconds. Is it appropriate to do the same if your agent is transacting from your account?

3. How would your future supreme agent react if it is asked about billing and shipping address while getting access to a paid article?

We have a saying internally– When the going gets tough, GDCs (Generation Defining Companies) get going! 

The road from human to agentic payments is bumpy but the size of the prize is as big as it gets. Solving agentic payments unlocks payments volumes and profits of all online// card not present transactions. This creates not just a market, but a once in a lifetime movement. 

This translates into ~$2.5 Trillion in global payments revenue and new age peers of generational payments companies created in the last decade. Think,

1. Honey ($4 bn cash acquisition by PayPal) for agents to figure the best offer codes. Honey is in fact the first ever agentic consumer payments product!

2. Lending compute to agents that create value and share profits 

3. Loyalty and consumer cards revenue moving away from issuers towards agents since consumers can have infinite reward programs and choose programmatically at the point of purchase

4. Merchants protecting against frauds of agentic purchases via KYA (Know Your Agent)

5. Enabling merchants to dynamically price their agents depending on factors like compute and demand

In this article we dive deeper into this generational shift from human to agentic payments.

One movement, Three phases

Source: EQT Ventures

Phase 1: Humans paying on AI interfaces

This is the first phase that will soon be triggered by the step change in discovery interface enabled by LLMs. Think about placing an order over the chat interface on Amazon.

Source

Phase 2: Agents paying on human interfaces

Source, Fun fact- Just reading the HTML of a doordash ordering page takes $1.87!

This is your assistant ordering lunch to the office on Doordash via RPA. This will be triggered by Consumer applications and B2B applications automating user workflows. We believe we will see agents paying for services sooner than products given quicker adoption of virtual receptionists and lesser variables than product purchases, but the jury is still out there.

Phase 3: Agents paying on agent friendly interfaces (all services on MCP)

Personal assistant ordering lunch via Doordash agent, food order agent paying weekend offer agent $0.01 to access latest deals. We believe this will kick in after service providers see early promise from agentic demand and the risk of not serving agents efficiently. Providers will then move quickly to offer agentic interfaces.

Emerging opportunity set

Deep dive: Phase 1, Humans paying on AI interfaces (mainly chat both text and voice)

Source: EQT Ventures

Our initial read is that the infrastructure required to support this phase is MCPs offering payment support that convert payment requests to right checkout nudge and CTA. Value will accrue with existing players like Paypal and Stripe that provide headless checkout over AI interface via chat and/ or voice.

Deep dive: Phase 2, Agents paying on human interfaces

Source: EQT Ventures

This is where we see more opportunities emerging.

Commerce RPA

Agents carrying out purchase steps as an RPA: this blends into agents RPA but doesn’t translate to a specialised fintech opportunity so we won’t go deep here

Agent friendly payment instruments

Low risk instruments that can be trusted to agents. Ex. Stripe single use virtual cards for agentic use cases, Payman, Nekuda, Coinbase Agentkit that propagates “every agent deserves a wallet”. Catena is building a bank for agents ground up. 

We are looking at,

1. Instruments with highly defined scope over parameters like amount, agent, merchant, frequency

2. Instruments that can be cancelled quickly at slightest risk

3. These instruments need to redefine risk as we know it. For instance, any traditional instrument would be banned by 3 transactions in 5 seconds but agentic instrument shouldn’t be penalised for quickly buying items on your monthly grocery list

Instrument optimisation

Third party services that help agents decide most optimum transaction instruments to optimise rewards, credit (liquidity) and success rate

Know your Agent

Agentic identity for compliance/ KYC/ AML: these registries are very nascent right now but will compound with scale. Ex. Cheqd, Nanda registry from MIT Media Labs

Orchestration

Orchestrators and PSPs for agents: low touch integrations that help agents integrate instruments, authorisation, transaction rules into payments to ensure high success rates. Who is building the Payrails for agents?

Additional UI for existing checkout providers, but hard to build independent opportunities

This category will see use cases like,

1. UI nudge to step up transaction: this will likely be supported by checkout providers together with agent provider apps

2. Post order communication: in this phase agents will maintain status of orders by leveraging a combination of scraping and APIs

While Visa and Mastercard support scoped tokens that can be used by agents, they unfortunately don’t yet support this phase since existing checkout flows do not accept tokens

Deep dive: Phase 3, Agents paying on agent friendly interfaces

Same as Phase 2, Phase 3 will need building blocks like agent friendly instruments, instrument optimisation, agentic identity and payments orchestration.

Following new opportunities are bound to emerge in this phase.

Middleware at service providers

Service providers would need low/ no code solutions to enable agentic payments. Protocols like x402 provide some out of the box but merchants will need to build much more complex flows that accommodate workflows including (but not limited to) dynamic pricing, offers, agent/ user communication.

Companies taking a crack at this include Paid AI (our portfolio), Visa, Mastercard, Checkout providers, Skyfire

Stablecoin facilitators for programmable payments

We might see stablecoin facilitators that blend with agentic instruments and middleware: this is contingent on stablecoins adoption by infrastructure providers. We are big believers in stablecoins but acknowledge that users primarily care about convenience, credit and rewards. They will have to be onboarded to stablecoins without knowing about it in the transaction journey. 

Call for startups!

We understand that the category is very new and evolving everyday. We are hungry to learn and help build Generation Defining Companies in this category. 

If you have gotten this far, we should definitely chat!

[Appendix 1/3] Zooming out, agentic payments enables new application businesses

1. Micro payments use cases will take off due to enhanced UI, ex. Pay $5 to read an article or add it to Perplexity knowledge base, paying $0.1 to add a song to a cafe playlist

2. Companies that help content be available to AI- Startups like linkup.so and Human Native can have their moment now that agentic payments interfaces become a reality 

[Appendix 2/3] Other second order effects

1. If you think about it, payment information is just one piece of information that the account can hold. If users trust an agent with a wallet ID they can also share their other details like address and preferences which agent can use to place orders. This could be a wave of users adding more and more information on chain after adopting agentic payments.

2. API calls coming from addresses with an attached wallet can be taken more seriously from the servers. [bad example I know but there has to be something] This can lead to high trust engagement like no captcha. 

3. API security will become a consumer product

[Appendix 3/3] x402 is elegant and promising

While we admire the elegance of the x402 protocol, we acknowledge that other protocols might eventually win. We are bullish because, 

1. Ridiculously easy to get up and running on (ref. tutorial)

2. Has biggest distribution on internet: HTTP

3. HTTP 402 logically and intuitively feels like the most optimal and intuitive transition from internet to agentic economy. Any other solution will have a more complicated call graph.

Source

We feel facilitators will likely support other payment networks (ex. Master/ Visa/ AMEX) along with blockchain

Imagine you are building an app “Plan my outfit” [client] that checks the weather from service provider “Accurate weather” [server]

1. Client calls server API to check weather

2. Server responds saying this is not free, feeling good through the day is worth $0.01

3. Client attaches a signed packet authorising the payment of $0.01 on the completion of service and sends an API call back to the server

4. Server checks with “Facilitator” if the transaction is legitimate

5. If legitimate, server figures weather details

6. Once weather details are figured successfully, server asks the facilitator to carry out the transaction and collects the money 

7. After collecting money, server gets back to client with the weather details