World of Workato 2026: What Stood Out in Las Vegas

Boston SoftDesign attended World of Workato (WOW) 2026 at The Cosmopolitan in Las Vegas, September 22–24, from the product keynote through the AIRO breakout sessions. As a Workato partner delivering integration, automation and AI projects, we focused on one question: what changes for our clients?

The headline: Workato introduced its Control and Execution Platform for enterprise AI, a single layer to govern AI and put it to work across business systems. Nearly every announcement at the event builds on it.

The problem: AI at work is a tangle

The keynote opened with a picture most IT leaders will recognize: ChatGPT, Claude, Copilot, Gemini and a handful of agent builders, each with its own lines into SaaS apps, data, productivity tools and legacy systems. Each of those tools brings its own identity, security, governance, observability, connectivity and orchestration. Some of it is strong inside its own ecosystem, and most of it doesn’t carry across.

We hear the same thing from clients. The question used to be “which AI should we buy?” Now it’s “we already have four, so how do we govern them and get them to do real work in our systems?”

Workato framed this as the intent-to-outcome gap: AI spending rising steeply while business results stay flat. Its three-step answer set the structure for the rest of the keynote: control the chaos, execute at scale, and prove enterprise ROI.

Takeaway 1: one platform to control and execute enterprise AI

The new platform has three layers:

• Agents: discover, build, deploy and operate multi-agent solutions.
• Control plane: govern every model, agent, workflow, API and MCP server with central identity, policy, observability and cost controls, and route work across models without vendor lock-in.
• Execution plane: deterministic orchestration across agents, MCP servers, data, apps and APIs; a context graph of your data, apps and processes; connectivity to thousands of systems; and one runtime with no infrastructure to manage.

The platform is also model-neutral: Claude, ChatGPT, Copilot, Gemini and Agentforce can all sit on top of the same governed layer. For mid-market companies that don’t want to bet everything on one AI vendor, that matters.

Takeaway 2: the Enterprise AI Control Plane gives AI an owner, a budget and an audit trail

For most CIOs, this is the announcement to watch. The control plane bundles several pieces that are usually bought separately:

• AI Registry: one living inventory of the agents, models, MCP servers, tools, owners and policies in the company.
• Gateways for models, MCP servers, agents and APIs: one governed front door instead of every agent wiring itself into systems its own way.
• AI cost controls: token-level attribution with budgets, quotas, rate limits and chargeback, so every AI cost has an owner, a unit and a ceiling.
• Guardrails: harmful content, prompt injection, sensitive-data handling and denied topics, with violations logged and sensitive data rejected, redacted or tokenized.
• Agent Evals: turn flagged conversations into repeatable tests and check task completion and groundedness before an agent ships.

The Model Gateway got its own announcement: route every request to the right model automatically, based on cost, quality, task complexity, latency, availability, geography and data sensitivity, with load balancing and failover. In practice, you keep the models you use today and can swap them next year without rewriting your agents.

We built Ally Platform on the same idea for the employee side: people use models from OpenAI, Anthropic, Google and others in one governed workspace, with sensitive data masked before it reaches a model and every request audited. Seeing Workato apply that principle to agents and integrations confirms where the market is heading: governance and model choice belong in one layer you own, not inside each vendor’s suite.

Takeaway 3: model choice doesn’t make agents reliable, architecture does

The most useful part of the week for practitioners was a live demo of a patient-scheduling agent. Given loosely defined tools, the same request produced several different outcomes across runs. Given one intent-based tool that ran the multi-step logic inside the platform, it produced the same structured result, with an audit trail, every time.

Would a cheaper model fix the cost? Only a little. Swapping the model changes the price per token; the architecture decides how many tokens you use. Four principles came out of it:

1. Intent-based tool design. One composable tool such as “schedule a specialist” replaces a string of granular API calls.
2. Lean context windows. Agents see only the tools relevant to their role. Showing 200 tools when 8 are needed wastes context on every call.
3. Platform-side orchestration. Multi-step logic runs in the platform, not in the agent loop, and the model sees one clean result.
4. Structured outputs. Tools return typed outcomes such as scheduled, denied or pending, not raw payloads.

Workato packages this as Enterprise Skills: reusable, governed business capabilities, such as processing an invoice or verifying an employee’s identity, that agents call instead of improvising. That is how we approach AI invoice automation for NetSuite: AI reads the invoice and matches it to the right vendor and PO, while Workato recipes handle approval routing and bill creation in NetSuite on fixed rules, so the AI never improvises a payment.

Takeaway 4: AIRO changes how Workato solutions get built

AIRO is Workato’s new building experience and the centerpiece of the redesigned Workato product. Instead of clicking through wizards or prompting a copilot step by step, you describe a goal and a team of specialized agents (architect, builder, tester, operator, optimizer) works on it across the lifecycle: build, test, deploy, operate, observe, measure and improve.

AIRO has five parts:

• AIRO Foundry: the team of agents, choosing the best-suited model for each step.
• AIRO Memory: context about your processes, systems, roles and permissions that grows over time.
• AIRO Playbooks: your standards and architecture patterns, applied by default.
• AIRO Acumen: monitors what’s running, surfaces issues and recommends next steps.
• AIRO Connect: brings AIRO into Claude Code, Codex, Cursor and other MCP-compatible tools.

The breakout sessions were more concrete than the keynote. The capabilities that stood out: bring a problem instead of a spec, show AIRO what’s broken and let it work out the fix, and see what will break before you change a recipe. That last one, impact analysis, is where integration projects usually lose time.

Underneath AIRO sits Enterprise Process Context: a knowledge graph of customers, orders, contracts, employees and documents, plus a Live Process Graph of how work actually flows. What sets AIRO apart isn’t the LLM it uses but the platform underneath it. For anyone who has inherited hundreds of undocumented recipes, a live map of what talks to what may be the most practical feature of the week.

Takeaway 5: not every process needs an agent

For all the agent announcements, the more grounded conversations at WOW came back to the same point: agents are moving from the reasoning side of work (chatbots and assistants that research and recommend) to the execution side (agents that act, with approval, on core processes). That’s where the business results are, and also where the risk is.

Two things we expect from here:

• AI gateways will consolidate. Separate API, model, MCP and agent gateways are too many front doors. They will converge into a single AI control plane, which is the bet Workato just made.
• Integration stays the foundation. An agent is only as good as the systems and processes it can reach. Classic integration and automation remain the on-ramp, and agents earn their place where they’re cost-justified.

What this means if you run Workato or are evaluating it

Here’s what we’re recommending to clients after WOW:

1. Inventory your AI before you add more. List every assistant, agent, model key and MCP server in use, with an owner for each. That’s the AI Registry idea, and it pays off even before you buy anything.
2. Pick one process and move it from reasoning to execution. Invoice processing, employee onboarding and case intake are good candidates: high volume, clear rules, measurable results.
3. Design tools around outcomes, not endpoints. Wrap multi-step logic in governed recipes or Enterprise Skills that agents call, instead of exposing raw APIs.
4. Give AI spend an owner. Track token cost per process and set budgets now, while volumes are small.
5. Keep model choice open. Route through a gateway or workspace you control, so switching models next year is a configuration change, not a rewrite.

The next World of Workato is already set for May 18–20, 2027, at Resorts World Las Vegas.

If you want to map any of this to your own stack, whether that’s Workato, NetSuite, Salesforce or a multi-model AI rollout, talk to our team or see our automation and integration services.

Frequently asked questions

When and where was World of Workato 2026? WOW 2026 took place September 22–24, 2026, at The Cosmopolitan of Las Vegas.

What is the Workato Enterprise AI Control Plane? A governance layer for enterprise AI that combines an AI Registry; gateways for models, MCP servers, agents and APIs; intelligent model routing; token-level cost controls; guardrails; and agent evaluations.

What is Workato AIRO? AIRO is Workato’s agentic building experience: a team of AI agents that designs, builds, tests, deploys and operates Workato solutions from a described goal, with memory of your environment and your standards applied by default.

Is AIRO or the control plane required to use AI with Workato today? No. Workato recipes, agents and Enterprise MCP already let AI act on your systems. The control plane and AIRO add governance, cost control and faster delivery as usage grows.