Blog · Published · Updated
A2A (Agent2Agent): what it is and when a company needs it
Short answer
A2A, or Agent2Agent, is an open protocol through which AI agents from different companies find each other, describe their skills and hand each other tasks. Google announced it in April 2025, the Linux Foundation took it over in June 2025, and the stable version 1.0 came out on 12 March 2026. MCP connects an agent to tools. A2A connects an agent to another agent.
How we workDetails ↓
- 01Call30 minutes with an engineer, free
- 02PrototypeA working system on your process and data, free
- 03ImplementationTarget and fixed price in the contract; 50% at signing
- 04Acceptance50% once the agreed target is met
- net per agent for one process, paid once
- from €3,000
- from signing the contract to acceptance
- 6–16 weeks
- free prototype on your cases
- €0
We belong to the OpenAI and Anthropic partner programmes and learn through their official courses. What this means ↓
Author: Syntalith · 3 min read
Statistics
AI agent protocols: statistics 2026
1.8 mo.
payback of a €3,000 wholesale agent that answers 600 enquiries a month from customers' agents through A2A, in place of a salesperson at 6 minutes an enquiry. Scenario.
8
According to the A2A community (2026), version 1.0 of the protocol came out on 12 March 2026, and a technical committee with representatives of 8 large technology companies leads its development.
100+
According to the Linux Foundation (2025), more than 100 technology companies supported the A2A protocol, created by Google in April 2025, when the foundation took it over on 23 June 2025.
Source: Linux Foundation, launch of the Agent2Agent Protocol Project (2025) ↗
130
According to Gartner (2025), only about 130 of the thousands of AI agent vendors offer real agentic solutions.
40%
According to Gartner (2025), over 40% of agentic AI projects will be cancelled by the end of 2027 because of costs, unclear value or weak risk controls.
Figures checked: 22 October 2026
How A2A works
Every agent that supports A2A publishes an Agent Card: a JSON file with its name, description, service address, required authentication and list of skills. A public agent usually makes it available at /.well-known/agent-card.json on its own domain (agent discovery).
- 01Discoverythe customer's agent fetches the Agent Card and checks whether the other agent has the skill it needs
- 02Taska request for a quote for 40 units delivered to Poznań
- 03Workthe wholesaler's agent checks stock, the price list and the discount rules on its own side
- 04Resultthe answer comes back at once or after a few hours, with a history of states
The A2A 1.0 specification provides JSON-RPC, gRPC and HTTP+JSON bindings, and the client states the protocol version in the A2A-Version header. The version 1.0 announcement lists support for many tenants on one endpoint and signed Agent Cards, which let you check an agent's identity before the conversation. A technical committee with representatives of 8 large technology companies runs the protocol under the Linux Foundation.
A2A vs MCP
| MCP | A2A | |
|---|---|---|
| What it connects | an agent with tools and data | an agent with another agent |
| The other side | carries out a command, for example a catalogue query | plans its own work and may answer later |
| Example | the agent reads the stock level | the customer's purchasing agent asks the wholesaler's agent for a quote |
The protocols complement each other. The wholesaler's agent talks to the customer's agent through A2A and reaches its own warehouse through MCP. The article on MCP covers the second protocol in more depth. Google also designs commerce protocols to work with A2A: the Universal Commerce Protocol lets a shop expose its functions through an API, MCP or A2A (Google Developers Blog).
Where A2A makes sense for a company
The most interesting cases are B2B relationships with a repeated exchange of information. A wholesaler answers the agents of regular customers about availability, contract prices and delivery dates, and passes unusual enquiries to a salesperson. A manufacturer answers distributors' technical questions from its current documentation. These are illustrations without implementation results.
- Today
- 600 enquiries × 6 min = 60 h (€2,094)
- With the system, a person checks the output in 20% of the time
- €419
- Model, 3 calls per case, Claude Sonnet 5.5
- €16
- Cost of one case
- €3.49 today, €0.72 with the system
- Saving
- €1,659 a month
Payback of €3,000
1.8 months
The volume, the time and the checking share are the scenario's assumptions. €34.90 an hour is the EU average hourly labour cost for 2025 according to Eurostat. Model: 3,000 input and 400 output tokens per call at Anthropic's rates, converted at the ECB rate of 1.1205 USD per euro on 22 October 2026.
This count assumes there are agents on the other side that already send enquiries. If there are none, an A2A endpoint brings no traffic, and the same agent can handle enquiries from email, as quote automation describes.
Security and the decision
- Which partner or platform has an agent your agent will talk to
- Which task they will pass to each other, and how often
- Who on your side approves commitments: prices outside the price list, delivery dates
An agent that accepts tasks from outside must know who the sender is and what it may do. The Agent Card sets the authentication, and you set the business rules yourself: amount limits, the list of customers, operations that need a person and a record of every decision. Treat content from another agent as input that may contain prompt injection. We build agents that work with partners' systems as part of AI agents, and in the free process review we work out whether you need A2A today.
Worked example · Scenario
How we count the payback: answering requests for quotes
We run the same count for your enquiries in the free process review.
€2,094
a month for preparing 120 quotes from customer enquiries by hand
Payback: 1.7 mo.
30 minutes per quote from an enquiry and checking in 15% of the old time are our assumptions. €34.90 an hour is the EU average hourly labour cost for 2025 according to Eurostat, for companies with 10 or more employees. Model cost: Claude Sonnet 5.5, about 4,000 + 1,500 tokens per draft quote, at the ECB rate of 1.1205 USD per euro on 22 October 2026. The extra margin is a scenario: we assume that 1 in 40 enquiries quoted the same day ends in an order that a slow reply loses today, with a margin of €344. In the review we count it on your sales process.
See the full calculation →
