Changelog

We re-research listings, add new tools, and correct facts continuously rather than on a fixed release schedule, which makes it easy for something genuinely useful to slide past a reader who last checked in a few days ago. This is a plain recap of everything that changed between July 20 and July 23, 2026 — four new listings, six category rankings, two vendor pricing corrections we caught, and six new research pieces — each with one line on why it matters. We’ll keep publishing these as the index keeps moving.

New agents added to the index

• GC AI (research-agents, July 22) — an AI platform built specifically for in-house corporate legal teams, whose Research Agent runs parallel, cross-checked research across 13M+ US case law opinions, statutes and regulations. Worth knowing if you’ve only seen general-purpose research agents: legal research has its own citation-verification requirements a horizontal tool won’t meet.
• LiveKit Agents (voice-agents, July 22) — an open-source (Apache 2.0), self-hostable voice-agent framework from the infrastructure company that powers OpenAI’s ChatGPT Advanced Voice Mode. The one genuinely open-source, provider-agnostic entry in a category otherwise dominated by closed, hosted platforms.
• Rox (sales-marketing-agents, July 22) — an AI-native revenue operating system built on “agent swarms” (one persistent agent per account) that continuously research and update your existing CRM, with a real free self-serve tier (2,000 agent actions/month) in a category where most competitors are sales-call-only.
• Cresta (support-agents, July 22) — a contact-center platform, spun out of Stanford AI Lab, that unifies an autonomous AI Agent, real-time human Agent Assist, and Conversation Intelligence on one shared conversation record instead of three separate products. Enterprise, custom-quoted, no self-serve tier.

Six category rankings we shipped

We published a dedicated “Best AI Agents” ranking for every category we track, each with its own comparison criteria rather than one generic list stretched across every use case: coding agents, agent platforms, sales & marketing agents, research agents, voice agents, and support agents. If you’ve only used our all-category best-ai-agents page before, the category pages rank a smaller, more comparable set against criteria specific to that job.

Pricing changes we caught before they were obvious

Both of these are cases where a vendor’s sticker price stayed put but what it actually buys quietly changed underneath it:

• GitHub Copilot replaced its flat “unlimited” framing for Chat, agent mode, code review, the cloud agent, CLI and Apps with a monthly “GitHub AI Credits” allowance (Pro $15, Pro+ $70, Max $100 — only code completions and next-edit suggestions stay uncapped). Sticker prices ($10/$39/$100) didn’t move.
• n8n grew its Cloud AI Assistant credit allowance roughly 45–90x (Starter 50→2,300/month, Pro 150→up to 13,700/month) with its €20/€50 sticker prices unchanged.

Both are folded into our running pricing-shifts tracker, which now covers nine tools that changed how they charge in 2026, not just how much.

New research and guides

• Which AI Agents Won’t Tell You What LLM Powers Them? — we classified all 44 listings by how much each discloses about its underlying model. 10 of 44 (23%) name nothing beyond “proprietary” or “managed, multi-model,” concentrated almost entirely in support and sales-agent categories.
• What Is MCP, and Which AI Agents Actually Support It? — a plain definition of the Model Context Protocol plus a real count of documented support across our index (34 of 44, up from an earlier count of 15 once we re-checked each listing directly).
• How Much Do AI Agents Actually Cost? A Pricing FAQ — five widely-used agents (Fin, n8n, Clay, Lindy, LangChain), five genuinely different pricing shapes, transcribed from each vendor’s live page.
• The State of AI Agents 2026 data report regenerated against our current 44-tool corpus (up from 40).
• AI Agent Statistics 2026 — sourced adoption, production and trust figures from Gartner, McKinsey, LangChain and Stack Overflow, alongside our own index.
• Using LangChain tools inside CrewAI agents — the working integration pattern straight from CrewAI’s own docs, for anyone treating the two frameworks as an either/or choice when they’re not.

If you want the next one of these in your inbox instead of finding it here, the newsletter signup below covers new listings, category refreshes, and anything else we publish.