Blog / Copilot
Every major AI platform can now build an agent without code. What business teams need is an environment where that agent can actually do work.
A year ago, building an AI agent without code was a differentiator. Today it is a baseline. OpenAI now directs natural-language agent building to Workspace Agents in ChatGPT, and it recommends its Agents SDK for code-first workflows. Google sells no-code agent creation directly to business leaders through Gemini Enterprise.
That changes the question business teams should be asking. The question is no longer whether they can build an agent without writing code — they almost certainly can. The question is what the agent can do once it exists.
The hard part of agentic AI was never the creation screen. Anthropic’s 2026 State of AI Agents Report surveyed more than 500 technical leaders in late 2025. Their top obstacles were integration, data, and change — each about what surrounds the agent, not the agent itself.
Source: Anthropic & Material, The 2026 State of AI Agents Report — survey of 500+ technical leaders, late 2025.
Gartner’s forecast points the same way. It expects more than 40% of agentic AI projects to be canceled by the end of 2027 because of rising costs, unclear business value, or inadequate risk controls — and it warns about “agent washing”: assistants and chatbots rebranded as agents without real agentic capability.
An agent that can’t reach your systems, act on your data, or answer to your policies is a chatbot — whatever the platform calls it.
The requests business teams bring to an agent are rarely about conversation. They are about work — and every one depends on a live connection, the right person’s access, and a record of what happened.
A builder on its own
”Here’s a draft. Copy it into Outlook, add the recipients, and send it when you’re ready.”
No connection to the mailbox
No record of what was sent
A builder inside an environment
Copilot’s agent builder begins where the business owner already is: with the job to be done. A guided conversation walks through each stage while a live draft of the agent builds alongside it, ready to review at any point.
Shape the mission
Define the work
Capabilities & integrations
Tune the voice
Research the craft
Write the system prompt
Launch
The system prompt still exists, but it is written through that conversation — not a blank field filled in cold. Anyone who wants to refine it can open the agent editor, which also summarizes the agent’s capabilities, knowledge files, and integrations. Changes apply everywhere the agent is available.
The business owner decides what the agent is for, what it draws on, and how it behaves. None of that requires learning prompt syntax, and none of it should.
This is where Copilot differs from a standalone no-code builder. Every agent created in Copilot joins an environment that is already connected to the rest of the business — managed centrally in Cockpit.
Excel, Gmail, Google Calendar, Google Sheets, Outlook
MCP servers let agents work with external services in real time
Register a spec and agents call the REST API directly
Set up once by an admin, or each person connects their own account
Stored for backend execution — never exposed to the model
Each integration also carries a plain-language description written for the agent, which it uses to decide when and how to use that tool. That knowledge lives in the environment — not rewritten into every prompt.
Build an agent in minutes. It sends the email, reads the sheet, and calls the internal API from its first run — no token, scope, or endpoint in sight.
Handing agent creation to business teams only works if the organization can still see and control what those agents do. That is Cockpit’s job.
Cockpit gives administrators control over agents, users, groups, access policies, integrations, credits, costs, and compliance — with a full audit log. Its Copilot overview brings cost, insights, and an action queue for pending approvals into a single dashboard.
As we argued in Enthusiasm Is Not a Strategy, governance, properly built, is what enables scale. Business teams get the freedom to build; IT gets the visibility to say yes — the opposite of the conditions described in Shadow AI Is Not a People Problem.
Not a dumbed-down developer console. The no-code promise is only real if every screen holds to it.
Business teams are a distinct audience with a distinct purpose, and they should not be defined by the skills they don’t use. Designing for them means building around the decisions they already make every day: what outcome they need, which systems are involved, who should approve what.
That standard can’t stop at the builder screen. If connecting a tool, granting access, or reviewing what an agent did pulls the business owner into developer-shaped language, the promise breaks exactly where it matters.
Building an agent without code is no longer a claim any single platform can own. What still separates platforms is what the agent can reach once it exists, what it can safely act on, and how much of that a business team can use without an engineer in the loop.
No-code gets the agent built. The environment gets the work done.
Sources
Next step
Build agents in the language of the job and connect them to the systems your teams already use, with every action governed from Cockpit.
