who are the big 4 ai agents?

where the phrase comes from

Nobody at a standards body ever named a big 4 of ai agents. The framing is lifted from the big 4 accounting and consulting firms, and it got attached to ai because people want a short list they can memorise before a meeting. Search results will happily give you one, and the four names change depending on which page you land on.

The stable reading is the four labs whose models and agent frameworks show up in almost every comparison:

Some lists swap microsoft for amazon, or add meta because of open weights. Both substitutions are defensible, which tells you how soft the category is.

why the list answers the wrong question

Every name above is a model provider first. If you are shopping, the question you actually have is "which agent runs my work", and the answer to that is a different market: the platforms built on top of those models. Coding agents, customer support agents, and agent teams are largely built by companies that buy their models from the four above.

So the useful comparison has two axes, not one:

1. Who makes the intelligence. Openai, anthropic, google, meta, mistral, deepseek. This decides quality, latency, and price per token.
2. Who makes the workplace. The platform that gives the model tools, memory, permissions, and a place to run. This decides whether the work actually lands.

Confusing the two is how people end up comparing a model api against a product and concluding the model is cheaper. It is cheaper. It also does not do anything on its own.

the single-agent ceiling

Every big-4 agent product ships as one agent with a tool belt. That design tops out on work that needs more than one perspective. A single coding agent writes the change and reviews its own change, which is exactly as reliable as it sounds. The alternative is a team: a planner that breaks work down, specialists that do it, and a separate reviewer whose only job is to disagree.

That is the split worth caring about when you compare vendors, more than which four logos made someone's list.

how this works in aldena

Aldena sits on the second axis. It does not train models. It runs teams of them: eleven prebuilt roles including a project manager, engineers, and a reviewer, arranged in an org chart where managers delegate down their own lines.

You pick which model each agent runs on, from the same providers listed above, at the published rates. One room can mix them: a cheap model on the analyst, a strong one on the reviewer. If you want the list of who builds the models rather than who runs them, that is the major developers of ai.