Everyone Is Selling Agents Now. Are They Though?

The word "agent" has become the most overloaded term in enterprise software. Every vendor selling AI to hedge funds and asset managers has one. Most of them are showing you a chat window.

That distinction matters, because the infrastructure decisions investment teams make this year will either constrain or compound the returns on their AI investment for the next five.

What most vendors mean by "agent"

In a large number of current product demonstrations, "agents" are advanced search interfaces with a generative layer on top. You type a question, the system retrieves relevant documents, a large language model synthesizes a response, and the result appears in a chat interface. That is genuinely useful. It is also not an agent.

A true AI agent has four properties that set it apart from a chat interface. It perceives its environment without being asked, monitoring sources and autonomously detecting new events. It plans across multiple steps, breaking a goal into a sequence of actions and deciding which tools to call and in what order. It executes those actions independently, querying data sources, writing structured outputs, and passing results to subsequent steps without a human trigger at each stage. And it closes the loop, with the output of one step conditioning what happens next.

A chat interface with a "deep research" button does none of this on its own. It responds to you. An agent acts on your behalf, continuously, on a schedule you define.

Why this matters for investment research

The workflows where agentic AI delivers the most value in investment research are inherently multistep and time sensitive: monitoring filings across a portfolio of 200 companies and flagging material events before a portfolio manager arrives in the morning, extracting comparable metrics across 50 earnings calls within a single reporting season, running a systematic ESG audit across a universe of holdings against a bespoke scoring methodology. None of those tasks reduce to a single query in a chat window.

If the demo you are being shown does not include evidence of the agent acting autonomously, across structured data, over a defined workflow, without a human prompt at each stage, then you are looking at enhanced search. That is a meaningful product, worth buying on its own merits. Just do not pay an agent premium for it.

Five questions to ask any vendor

• Can the agent run on a schedule without a user initiating it? How?
• Can it produce structured data outputs, not just narrative summaries?
• Does it connect to your internal data and proprietary sources, or only the vendor's own content library?
• Can you inspect and modify the agent's logic, or is it a black box?
• Is the underlying data layer built specifically for this domain, or is it a general purpose corpus?

The answers will tell you quickly whether you are being sold an agent or a chat interface with a more persuasive name.

What real agentic infrastructure looks like

Orbit has been building on genuinely agentic foundations since the platform launched. The Agent Marketplace deploys production grade agents across use cases including earnings monitoring, filing surveillance, ESG data extraction, and regulatory horizon scanning. Each one runs autonomously against Orbit's data layer: 70 million documents per year, 75,000+ companies across 120 countries, with exclusive coverage of 5,500 China A-share names. Agents connect to client environments via MCP and API, write structured outputs into internal systems, and run on schedules that match investment workflows rather than waiting for a user prompt.

They are also genuinely customizable. CANDRIAM runs a proprietary ESG scoring methodology through an Orbit agent at scale, producing consistent outputs across thousands of holdings without manual intervention. That is what a real agentic use case looks like.

Next week, the award-winning Orbit platform is releasing something that takes this considerably further. We will share more very shortly. If you want to be among the first to see it, reach out to the team or register your interest below.

These are conversations that matter deeply to us at Orbit. If you are thinking through how agentic AI fits into your research or data infrastructure, we should talk.