From chatbot to agent: what the next AI wave means for mid-sized companies

In short: The next AI wave is about orchestration: instead of one chatbot or a single agent, several specialised agents now cooperate through open standards such as MCP and A2A. Gartner expects that by 2028 about 33% of enterprise software will include agentic AI (2024: under 1%), and 15% of day-to-day work decisions will be made autonomously (2024: 0%). At the same time, Gartner warns that over 40% of agent projects will be canceled by the end of 2027 — mostly due to weak governance, not weak technology. Anyone who wants to benefit from the next wave should build solid governance for a first agent before connecting several of them.

What sets the next AI wave apart from today’s AI agent?

The first wave was the leap from a chatbot that only answers to an agent that finishes a task on its own — we broke down what that means in practice here. The second wave goes a step further: not one agent per task, but several specialised agents that talk to each other and solve a more complex task together. For that to work, systems from different vendors need a shared language — which is exactly what two open standards gaining traction in 2026 provide. Anthropic’s Model Context Protocol (MCP) connects an agent to its tools and data sources. The Agent2Agent protocol (A2A), which Google handed over to the Linux Foundation in June 2025, governs how agents from different vendors negotiate with each other and split up tasks. Google itself describes the division of labour this way: MCP handles internal tool integration, A2A handles the external coordination between autonomous entities.

Why are companies now building networks instead of single agents?

Because many tasks are too complex for one agent alone, yet break down cleanly into steps. In practice this often looks like this: a first agent researches and gathers the relevant data, a second checks it against existing systems such as ERP or CRM, a third drafts the result — a person reviews and approves it. Each agent stays limited to its specialism, which makes it easier to test and control than a single system meant to do everything. This shift explains why, according to the 2026 Bitkom AI study, AI agents are — alongside software development and knowledge management — the fastest-growing field of use, on top of AI adoption among German companies more than doubling within a year, from 17% to 41%.

Isn’t this just the next hype cycle?

That concern is fair — and Gartner itself fuels it with a stark figure: over 40% of agent projects will be canceled by the end of 2027, due to rising costs, unclear business value or inadequate risk controls. Gartner analyst Anushree Verma sums up the core problem: most agentic AI projects today are early-stage experiments, mostly driven by hype and often misapplied. This is compounded by “agent washing” — existing chatbots and automation tools relabelled as “agents” without real autonomy behind them. With several networked agents, this risk multiplies: every additional agent is an additional point of failure, cost and control. But it is equally true that the standards themselves are no longer hype — A2A has been production-ready as version 1.0 since March 2026 and is backed by over 100 technology companies, including AWS, Microsoft, SAP and Salesforce. The difference between hype and real progress, then, lies not in the tool but in how disciplined the rollout is.

What does this mean for governance inside a company?

This is exactly where Gartner’s latest advice comes in — published as recently as 20 August 2026: the first agent pilot should be treated as a governance pilot, not an ROI pilot. Gartner analyst Alex Levine explains why: the greatest risks and the greatest value both lie in building oversight, not in chasing quick returns — most early pilots fail because of unclear controls, not poor technology. The recommendation is a low-risk, reversible process with clear boundaries: which data the agent may see, which actions it may take, and where human review is mandatory — all defined before development even begins. That includes clear ownership spanning the business unit, IT and audit, plus a sandboxed environment where the agent’s behaviour can be observed without exposing live operations. Success is then measured not by autonomy or ROI, but by traceability: can you reconstruct, for every run, what the agent planned, accessed and produced?

How should mid-sized companies prepare now, without rushing?

By getting the basics right before connecting several agents. A mid-sized company that already runs a single agent reliably, with clear governance, has solved the harder part — networking several agents later becomes an extension, not a fresh start. Concretely, that means:

• One agent first, the network later. One process, one tightly scoped role, one person who stays accountable — that foundation also carries the next stage.
• Favour open standards over closed systems. Solutions that support MCP and A2A can later connect to other agents and vendors without rebuilding the entire architecture.
• Governance before autonomy. Access rights, logging and a clear escalation path belong there from day one, not as an afterthought.
• Build in data protection from the start. In the D-A-CH region especially, this is not a brake but a selling point: agent workflows can be built so that sensitive data is anonymised or processed locally.

This is exactly where weooo comes in: not chasing the next agent trend, but setting up the one process cleanly enough that it can be extended later — GDPR-compliant, traceable, and built on open standards so no vendor lock-in creeps in. No buzzword bingo, just potential over hype.

What does this concretely mean for the next 12 to 24 months?

The shift is gradual, not sudden. 2026 is the year in which single agents become everyday practice in mid-sized companies, while the technical standards for connecting them stabilise at the same time. The real spread of multi-agent systems — Gartner’s 33% by 2028 — is still ahead of us, not behind us. For companies deciding now, that means: there is no reason to wait for the finished big leap, but equally no reason to rush into connecting several agents before the first one runs reliably. Whoever does the governance groundwork today will not be the one playing catch-up in two years — they will simply add the next stage on top.

Conclusion

The next AI wave is not a new hype term but a technical shift backed by solid standards: from single agents to coordinated agent networks. The opportunity is real — and so is Gartner’s warning that over 40% of projects fail once governance becomes an afterthought. The smartest path for mid-sized companies is therefore not to jump on the next trend, but to lay the governance groundwork now, while their own agents are still small enough to manage.

Want to know whether your first AI agent is ready for the next wave? Talk to us — we’ll look at your architecture and give you an honest read on what makes sense today and what can wait.

Sources

• Gartner, “Gartner Predicts Over 40% of Agentic AI Projects Will Be Canceled by End of 2027” (25 June 2025) — forecasts of 33% of enterprise software with agentic AI by 2028 (2024: <1%), 15% of autonomous day-to-day decisions by 2028 (2024: 0%), quote from Anushree Verma, “agent washing”, roughly 130 real vendors.
• Gartner, “Gartner Says CFOs Must Pilot Governance First Before Scaling AI Agents” (Q&A with Alex Levine, 20 August 2026) — governance pilot instead of ROI pilot, recommended criteria for a first agent pilot.
• Bitkom, “Künstliche Intelligenz in Deutschland” — 2026 AI study (representative survey of over 600 companies with 20+ employees): AI adoption 41% (2024: 17%), AI agents among the three fastest-growing fields of use.
• Google Open Source Blog, “A year of open collaboration: Celebrating the anniversary of A2A” (16 April 2026) — A2A protocol version 1.0 since March 2026, over 100 supporting companies (including AWS, Cisco, Microsoft, Salesforce, SAP, ServiceNow), MCP/A2A division of labour.