AI agents enterprise: from chatbots to accountable workflows
AI agents in the enterprise are software-driven workers that carry out multi-step business tasks across existing systems, combining domain context, workflow orchestration, and tool access to automate decisions and execution rather than simply answering questions like traditional chatbots. The most important shift is that AI agents enterprise strategies are turning messy, one-off consulting projects into repeatable, accountable software workflows. Instead of hiring more people to push tickets and documents, organizations are starting to encode how the work should be done and letting agents execute it at scale. That is a much bigger change than rolling out another conversational interface. It moves AI from experimentation on the side to the operational core, where indemnity margins, downtime, and compliance risk live. The winners will be the firms that treat agents as first-class workers, not toys bolted onto a helpdesk.
Guidewire’s agentic framework: insurance workflow automation with teeth
Guidewire’s Qusar release makes a strong argument that insurance workflow automation only works when AI agents live inside core systems, not on the edges. Its Agentic Framework lets property and casualty carriers build and manage agents directly on the Guidewire Cloud Platform, with secure, real-time access to policy, claims, and billing data and workflows. That matters more than any model benchmark, because underwriting and claims decisions depend on context and traceability. Guidewire is not selling a general-purpose assistant; it is shipping insurance-aware agents like claim summarisation and policy change assistants that collapse multi-day, multi-click work into minutes. As Celent’s Karlyn Carnahan noted, insurers see that AI value comes from deep integration with core business processes, not isolated experiments. In practice, that means agents that understand Gosu code, product configuration, and data curation, and can be governed like any other part of the platform.
This is what a serious agentic framework looks like: agents that carry real underwriting and claims load, while still fitting into existing controls and compliance structures. Anything less is a toy.
Kyndryl’s agentic modernization: infrastructure modernization AI as a service
If Guidewire shows agents changing front-office insurance work, Kyndryl shows how infrastructure modernization AI can eat the back office. Its Agentic Modernization services-as-software turn chunks of infrastructure and application modernization into prebuilt workflows delivered through Kyndryl Bridge. Discovery, code analysis, dependency mapping, system design, code generation, testing, and validation across mainframe, cloud, network, and distributed environments all become repeatable, AI-driven workflows instead of one-off consulting slides. Importantly, Kyndryl keeps oversight, governance, and final decisions with human teams, using AI agents as execution engines rather than replacement architects. That is the right balance: encode what is repeatable; keep judgment and accountability with people. The impact is not only lower effort per project, but a structural shift away from billing for effort toward selling outcomes packaged as software. In a market drowning in legacy estates, that is where modernization needs to go.
Agent Plugins 1.0: the missing portability layer for AI agents
All this agentic ambition collapses if every platform speaks its own language. The Agent Plugins 1.0 standard is the first serious attempt to give agents a write-once-run-anywhere container model. Vercel, Amazon, Cursor, Microsoft, and OpenAI contributed to a spec that packages an agent’s skills and tools in a single directory structure, building on the MCP protocol and the Agent Skills standard. Skills define how agents perform tasks; MCP connects them to databases, APIs, shell scripts, and other resources. Together they make capabilities modular and portable across clients like VS Code, Cursor, GitHub Copilot, ChatGPT, Codex, and Kiro. According to the Linux Foundation’s Agentic AI Foundation, the scope is deliberately narrow so the shared format stays predictable while still letting vendors experiment in a namespace. That restraint is good news for enterprises: portability requires boring standards, not flashy lock-in.
From consulting to accountability: what enterprises must change next
The common thread across these moves is that services firms and software vendors are quietly rewriting their business models around AI agents. Guidewire bakes agents into its core platform. Kyndryl turns recurring modernization work into reusable workflows. The Agent Plugins ecosystem pushes everyone toward shared tools instead of fenced-off stacks. That forces enterprise agencies and internal teams to rethink their role: less throwing bodies at tickets, more designing accountable AI-assisted software development patterns. The real competitive advantage will not be access to a better model; it will be owning high-quality agent workflows, clear governance, and portable tools you can run across platforms. Enterprises that cling to traditional consulting patterns will pay repeatedly for the same work. Those that treat agentic frameworks as core infrastructure will ship faster, with clearer lines of responsibility. AI agents are no longer an experiment. They are becoming the operating system for complex work.
