Back in 2024 (it seems so long ago now) I wrote about Agents (links below) and cautioned about how early we were in their evolution. Now, almost a year later we seem to be in a completely different place – brought back to my mind to revisit by the recent announcements from:
- Langflow – releasing v1.6
- Microsoft – consolidating AutoGen and Semantic Kernel into the Microsoft Agent Framework
- OpenAI – releasing AgentKit
(I still have concerns about the role of Unions in this new Agentic world – see LinkedIn: The Roles of Unions in an Agentic AI World)
Back in 2024 it seemed like everyone said something like “2025 will be the year of Agents.” Where are we now, seemingly surrounded by pilots, prototypes, and proof-of-concepts – but not the mass adoption (as Georg Zoeller recently replied to me in a post “…maybe the apocalypse will be postponed for another year, again.”). With the maturity we have seen in platforms and technology over the last year I’m not sure what we have seen in 2025 is failure, it’s more like fermentation. The tools are stabilizing, the standards are forming, and people and organisations are learning.
My take, 2026 could indeed be the year Agents start to take hold, but only for those who are laying the groundwork today.
What Is It?
It’s worth spending a little time revisiting the terminology.
- Agents & Agentic AI: Agents are autonomous (or semi-autonomous) systems that can react to their environment, reason about goals, plan steps, call tools or APIs, and adapt their actions based on feedback. They’re not just chatbots, they’re goal-seeking systems that can decide what to do next and have access to tools to get that job done.
- Workflows (vs. Agents): Traditional workflows follow predefined logic – fixed steps and approvals. Agentic workflows adapt, handle ambiguity, and collaborate with humans or other agents to achieve outcomes. (Although tools like n8n, Make, Zapier blur those lines by adding ‘Agent’ functions into their workflows.)
- Agent Platforms / Frameworks: These provide orchestration, safety, and integration. Examples include: Microsoft MAF, Copilot Studio, Azure AI Agent Service; OpenAI AgentKit; Langflow; LangChain; CrewAI and Google Agentspace.
- Industry-specific platforms: E.G. for customer service, research, and operations like Salesforce Agentforce
- Role vs Task: A way to think about the difference between an Agent and a Workflow (although like all analogies, if you stretch them they break) is to think of an Agent filling a role compared to a Workflow which performs a task, a sequence of steps.
What’s New in 2025 The focus seems to be shifting, the questions are evolving, from “can we build an agent?” to “can we govern and scale them?” with attention to robustness, observability, safety, and interoperability.
What Does It Mean from a Business Perspective?
Agent adoption now raises not just technical, but strategic decisions. Here’s what leaders should be thinking about:
Operational Efficiency – with Limits: Early adopters are automating reporting, customer triage, and research tasks. But efficiency gains only stick if agents are well-instrumented, governed, and continuously managed and tuned, there is a shift in the overhead costs, it’s not free.
The No-Code vs. Dev-Friendly Divide: Organizations face a key decision:
- No-code / Low-code platforms: (like Copilot Studio or other enterprise agent builders) promise faster experimentation, governance controls, and business-user accessibility.
- Developer-oriented frameworks (LangChain, Microsoft Agent Framework SDK, OpenAI Agent Kit) provide maximum control, extensibility, and custom integration.
- Blended approaches: which provide a range of options – think Microsoft’s environment where Agents span things as simple as a SharePoint Agent, through the Azure AI Agent Service to dev focused frameworks like MAF, or Langflow which while it has a GUI canvas for development lets you get to the underlying code or develop custom tools.
Early Adoption Risks: Jumping early with any technology introduces risk – agents are no different, including
- Platform Volatility: Platforms, APIs and SDKs evolve rapidly – builds may change overnight (not good when you are aiming for stability in your business processes).
- Hidden Complexity: state management, context persistence, and tool selection are non-trivial.
- Governance Gaps: poorly scoped agents can act outside policy or jurisdictional boundaries.
- Data Exposure: agents often process internal data, creating privacy and compliance risks.
- Human Trust & Accountability: when agents make decisions, who’s responsible if something goes wrong?
Strategic Advantage for Early Learners – having said this, particularly about risk, the first wave of adopters, those who combine agility with structured governance, will emerge in 2026 as the “Agent-ready” organisations with experience, patterns, and internal frameworks already in place.
What Do I Do With It?
Here’s how to prepare without overcommitting:
- Start with a Pilot and a Playbook Choose a single high-value, internal – not customer facing – use case (e.g., onboarding assistant, RFP summarizer, or automated research). Document the architecture, governance, and what was learned as a reusable playbook.
- Establish an “Agent Architecture” View Map out how agents will integrate with your data, APIs, and identity systems. Decide early whether you’ll pursue a low-code, developer-first, or hybrid approach. (Get your Enterprise, Solution and Data Architects involved.)
- Build Guardrails and Observability Early Instrument logs, audit trails, and approval workflows (‘human in the loop’ agent involvement in key steps) from the outset. These aren’t optional – they’re what makes agentic systems enterprise-grade.
- Cross-Functional Governance Involve compliance, legal, and security teams early. Their input now will save you from regulatory headaches later.
- Upskill for Agent Thinking Developers need to learn orchestration and reasoning patterns; business users need to understand prompt logic and oversight.
- Iterate and Learn Publicly Share internal findings, across departments – collective learning accelerates maturity. (Community of Practice anyone?)
One thing is certain, Agents are not a passing trend, they are a structural change in how we build and interact with software. 2025 wasn’t “the year of Agents” because the tooling, governance, and business readiness weren’t there yet. 2026 might be, just might be, but only for organizations that experiment wisely now, balancing innovation with control.