Agentic Landscape
The 2026 agentic-AI market, end to end. Every platform you'd plausibly consider: what it does, what it costs, where it fits, and what to watch out for. Verified against vendor pricing pages and independent reviewers.
01 · The market in one page
Six shapes, three pricing archetypes
The agentic-AI market has settled into six rough shapes. Most "agent platform" reviews bury this, so it's up here.
| Shape | What it is | Typical buyer | Pricing model |
|---|---|---|---|
| Enterprise agent suites | A vendor's existing SaaS with an agentic layer bolted on. | Big companies already on the suite | Per-action / per-conversation / per-credit (metered) on top of per-user |
| Developer frameworks | Code libraries for building agents. | Engineering teams shipping product | Open-source (free) + paid cloud/observability on top |
| No-code / low-code builders | Visual workflow tools that now ship with native AI agent nodes. | Ops, growth, IT, citizen developers | Per-workflow-execution / per-task / per-credit |
| Autonomous / browser agents | Computer-use agents that drive a real browser or desktop. | Individuals, devs, specific verticals | Per-month subscription + usage overage |
| Open-source personal-agent harnesses | An open-source runtime you install (or self-host) that runs a persistent personal agent with memory, sub-agents, skills, and chat-app distribution. | Power users, founders, indie hackers | OSS (free, BYO model) + optional managed cloud |
| Coding-agent IDEs | AI agent that lives inside (or replaces) an editor. | Software engineers | Per seat (tiered) + usage overage |
| Voice agent platforms | Conversational AI that picks up the phone: low-latency speech-to-speech with function calling, telephony in/out, and a turn-taking model. | ||
| Agent observability & evals | Tracing, evaluation, prompt management, and gateway logic for production agents. | ||
| EU AI Act compliance & governance | Tools that turn the EU AI Act into operational evidence: risk classification, runtime transparency, audit trails, drift detection. |
Three pricing-model archetypes you will see everywhere: per-action / per-conversation / per-credit: easiest to forecast, hardest to control. "Action" definitions vary wildly. Per-token: most transparent, but punishes long-context and multi-step agents. Per-seat / per-month: easy to budget, terrible for variable workloads, generally hides the model cost.
02 · Pricing-model archetypes
"Credit" ≠ "token" ≠ "action" ≠ "task"
These units are routinely conflated. They are not interchangeable, and a "credit" from Vendor A is rarely the same as a "credit" from Vendor B. The right way to think about them: a unit defines what you're really paying for, and that changes which workloads are cheap and which ones punish you.
| Unit | What you're really paying for | Volatility | Where it shows up |
|---|---|---|---|
| Token (in + out) | The LLM call itself, plus sometimes embedding/tool tokens | High: long context, retries, thinking tokens | LangSmith, AWS Bedrock, Google Gemini API |
| Action | A "thing the agent did": definition varies (one LLM call? one tool call? one full plan-execute cycle?) | Medium: depends on agent design | Salesforce Agentforce, AWS AgentCore, Manus, Replit Agent |
| Conversation | A full session between user and agent (regardless of internal steps) | Low: easy to forecast | Agentforce (legacy), Copilot Studio (legacy) |
| Credit / LCU | Vendor-defined unit, usually pegged to ~1k tokens of model work + tool overhead | High: opaque until you read the rate card | LangSmith LCU, n8n AI credits, Make, DevRev, Lindy |
| Task / ZAP / execution | One full workflow run, possibly multi-step | Low | Zapier, Make, n8n |
| ACU / compute unit | A bundle of compute, often tied to a specific model's cost | Medium | Devin ACUs |
| Per-seat / per-month | A human user license; agent usage is a separate meter on top | Flat | Almost all enterprise suites + consumer agents + coding IDEs |
Rule of thumb: if the vendor charges you for tokens and a flat seat, you are paying twice. If the vendor charges per action, ask them for the rate card before signing.
03 · The platforms
Six categories, 89 platforms
Click any category to expand or collapse. Use the filter at the top to narrow down by name, tagline, or use case. Each entry cross-references the ones it competes with.