Everyone else worth knowing

Not every name matters. These do, because you will meet them. Grouped by what they are for, with the one thing to know about each. September 2026.

Search and answers

Perplexity. A search engine that answers in sentences with sources instead of a list of links. Passed 100 million monthly users in 2026 and was valued at over $20 billion. The closest thing to "Google, but it just tells you."

Writing code (and, increasingly, everything else)

Cursor. A code editor with an AI agent built in, where a large share of new software is now written. Its cloud Automations can run agents on a schedule, which is why it is one of the five homes this site writes for.

Claude Code and OpenAI Codex. The two labs' own coding agents, run from a terminal or a desktop app. Claude Code is the other technical home this site supports.

Replit, Lovable, Bolt. "Describe an app, get an app." For people who have never coded. Fine for a prototype, and worth knowing that the result is real code you will eventually need someone to look after.

The open-weight labs

DeepSeek (China). In early 2025 it released a model nearly as good as the leaders at a fraction of the cost and briefly knocked hundreds of billions off Nvidia's value. Its V4 models in 2026 kept the pressure on. Free to download, cheap to run.

Alibaba's Qwen (China). A large, well-regarded open family that developers reach for constantly and consumers rarely hear of.

Mistral (France). Europe's main lab. Raised at a $20 billion valuation in June 2026, sells to governments and companies that want a non-American option, and publishes open weights for most of its models.

Cohere (Canada). Builds for businesses rather than consumers, with a focus on many languages and on running inside a company's own walls. In 2026 it absorbed Germany's Aleph Alpha.

Hugging Face. Not a lab, a library: the site where almost every open model is published and downloaded, nearly three million of them. Nvidia agreed to buy it in September 2026.

Automation and glue

Zapier and Make. The old way to connect apps to each other without code (when this happens in Gmail, do that in Sheets). Both now have AI steps, and both are being squeezed by agents that can do the connecting themselves.

n8n. The same idea, open source, popular with people who want to run it on their own machine.

MCP. Not a company, a standard, created by Anthropic and handed to an open foundation. It is why one connector works in Claude, ChatGPT, Grok and Cursor alike, and why "connect your AI to X" stopped being custom work in 2026.

Where agents get their hands

Robinhood. The first consumer brokerage to give AI agents a proper door: a separate agentic account, an MCP connector, and orders that must be previewed first. When this site adds agents that can place an order, that door is the one they will use, and the separate account is why they could never reach your main one.

Google Workspace and Microsoft 365. Gmail, Calendar, Drive, Outlook, Teams. The connectors most everyday agents need, and both companies now expose them to outside AI apps.

Notion, Slack, Linear, GitHub. The work tools with the most mature connectors. If your job lives in one of these, an agent can too.

Voices and pictures

ElevenLabs (voice), Midjourney and Black Forest Labs (images), Runway and Pika (video). Specialists that are better at one medium than the big labs, at least for now. Fun, and mostly outside what this site builds.

Two you can safely ignore

Character.ai and the companion-app crowd: chatbots with personalities. Big audiences, not much to do with agents. Any "AI trading bot" sold on social media: see staying safe. Every one of them.

What to remember

• Perplexity for search, Cursor for code, Robinhood for money, Google and Microsoft for the everyday connectors.
• The open-weight labs (DeepSeek, Qwen, Mistral) keep prices falling for everyone, including the labs you pay.
• MCP is the reason connectors are boring now. Boring is good.