Every organisation I speak to is building or evaluating AI agents. Almost none of them are building agents that can communicate, collaborate, or transact with each other across system boundaries.
That gap is not a minor technical inconvenience. It is the difference between AI that automates individual tasks and AI that transforms how entire businesses operate.
Three open protocols − now governed by the Linux Foundation with the backing of every major AI provider on the planet − are closing that gap. South African enterprises that understand them now will be in front. Those that discover them later will be rebuilding.
In the 1990s, the internet did not become the internet because computers got smarter. It became the internet because HTTP standardised how browsers talk to servers, and TCP/IP standardised how data moves between networks. Intelligence became useful at scale because communication became standardised first. We are at exactly that moment for agentic AI.
Gartner forecasts that 40% of enterprise applications will feature built-in AI agents by 2026 − up from less than 5% in 2025. IDC projects agentic AI spending will exceed $1.3 trillion by 2029. In my years advising enterprise technology leaders across telecommunications and financial services in Africa, I have never seen adoption forecasts this steep hold for technologies that remain locked in siloed, non-interoperable deployments.
The protocols that allow AI agents to communicate, collaborate and transact are not future infrastructure.
The protocols that allow AI agents to communicate, collaborate and transact are not future infrastructure. They are the foundation being poured right now, and the organisations pouring it will have a structural advantage that compounds over time.
Three protocols, three distinct problems
The confusion I most commonly encounter is treating these protocols as competing alternatives. They are not. They are complementary layers of a single stack, each solving a fundamentally different problem.
MCP − the Model Context Protocol, developed by Anthropic in 2024 and donated to the Linux Foundation’s Agentic AI Foundation in December 2025 − solves the agent-to-resource problem. An AI agent asked to check CRM for overdue orders needs a standardised way to access that data. MCP provides it: a universal layer between any agent and any external tool, database, API, or service.
Write one MCP server for your Salesforce instance, and every MCP-compatible agent can use it. By February 2026, MCP had crossed 97 million monthly SDK downloads and had been adopted by every major AI provider: Anthropic, OpenAI, Google, Microsoft and Amazon. NVIDIA CEO Jensen Huang called MCP’s adoption in late 2025 nothing less than a revolution in the AI landscape.
A2A − the Agent-to-Agent Protocol, launched by Google in April 2025 and now under Linux Foundation governance with over 100 enterprise supporters − solves the agent-to-agent problem. When your inventory management agent detects a stock shortage and needs to delegate a procurement task to a supplier agent built by a different vendor on a different platform, how does it do that?
A2A defines how agents discover each other via structured Agent Cards, delegate tasks, exchange updates in real-time and maintain auditable workflow logs. The list of A2A’s enterprise backers reads like the client list of a major consulting firm: Salesforce, SAP, ServiceNow, PayPal, Atlassian, and the strategy arms of Accenture, Deloitte, McKinsey and PwC.
ACP − the Agentic Commerce Protocol, co-developed by OpenAI and Stripe and released September 2025 − solves the agent-to-merchant problem. Until September 2025, there was no secure, standardised way for an AI agent to complete a purchase on behalf of a user. ACP fills that gap.
ChatGPT users can now buy directly from Etsy sellers and over one million Shopify merchants − including Glossier and SKIMS − without leaving a conversation. The Shared Payment Token at the heart of ACP is scoped to a specific merchant and transaction amount, never exposing raw card credentials, and powered by Stripe Radar with near-zero fraud rates since launch.
To summarise: MCP for context. A2A for collaboration. ACP for commerce. Not competitors − layers. Together they form the connective tissue of an economy where AI agents get things done on behalf of real people and real businesses.
Why the governance moment matters
The December 2025 launch of the Linux Foundation’s Agentic AI Foundation − co-founded by OpenAI, Anthropic, Google, Microsoft, AWS, Block, Cloudflare and Bloomberg − marks something that almost never happens in technology: direct competitors agreeing to build on shared infrastructure.
MCP and A2A are now housed under neutral governance with a platinum member list that includes the organisations building the tools that will run on top of these protocols. When that happens, a standard stops being a proposal and starts being the floor.
For South African enterprises, the governance structure matters for a specific reason. POPIA compliance requires that AI-driven workflows processing personal data be auditable and explainable.
A2A’s structured task lifecycle − with defined states, structured delegation logs and Agent Cards advertising what each agent is authorised to do − provides exactly the kind of traceable audit trail that POPIA accountability obligations demand.
An agentic workflow built on A2A is not just more capable than a custom integration. It is more auditable by design.
What South African enterprises should do now
The practical starting point is not A2A or ACP − it is MCP. MCP is the most mature, most widely adopted and easiest to deploy. Auditing core data systems − CRM, ERP, core banking, regulatory knowledge bases − and building MCP servers for the resources AI agents most frequently need is the single highest-leverage near-term investment.
It pays dividends immediately and positions the organisation for A2A-enabled multi-agent orchestration as that capability matures.
For financial services and telecommunications organisations, the compliance review chain is the most compelling first A2A use case. A lead agent receives a compliance query, delegates jurisdictional analysis to a regulatory knowledge agent, delegates client history retrieval to a CRM agent, and synthesises a structured, auditable response − all without custom integration code between any of those agents.
Every delegation, every state change, every response is logged in A2A’s structured task format. That is not just efficiency. That is the audit trail regulators are going to ask for.
The window is open
The analogy I keep returning to with clients is the early internet. In 1993, most organisations were asking whether to build a website. The organisations that were asking how to build web infrastructure − how to connect their systems, structure their data and participate in the emerging standard − are the ones that defined the next decade of commerce.
We are at the same inflection point for the agentic internet. The protocols are real, adopted and governed by a coalition with no meaningful incentive to fragment.
My question for every South African technology leader: when your competitors’ AI agents can discover your agents, delegate tasks to your systems and transact on behalf of shared customers without a single custom integration − will your organisation be a node in that network, or will it be invisible to it?
Sources:
- Anthropic MCP Documentation (2024) − modelcontextprotocol.io
- Linux Foundation − Agentic AI Foundation launch, December 2025, linuxfoundation.org
- Google A2A Protocol − github.com/a2aproject/A2A (April 2025; 100+ enterprise supporters by Feb 2026)
- Stripe / OpenAI − Agentic Commerce Protocol, September 2025 − agenticcommerce.dev
- Gartner (2025) − 40% of enterprise apps will feature AI agents by 2026 — gartner.com
- IDC (2025) − Agentic AI spending to exceed $1.3 trillion by 2029 at 31.9% CAGR − idc.com
- Jensen Huang / NVIDIA (Nov 2025). MCP has revolutionised the AI landscape
- Microsoft Azure Blog (May 2025) − empowering multi-agent apps with A2A — microsoft.com/cloud-blog
* Eugene Perumal is a strategy and architecture principal with over 20 years' experience in enterprise technology across telecoms and financial services, including senior roles at Vodacom Group and Absa Group. He holds Master’s degrees and certifications in enterprise architecture, AI governance, cloud and analytics. He writes on enterprise AI strategy, ROI measurement and the shift to agentic AI deployment.
Share
