KMM Technologies
Home
ReferenceLast updated 2026-06-17 · v3.189 platforms

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.

ShapeWhat it isTypical buyerPricing model
Enterprise agent suitesA vendor's existing SaaS with an agentic layer bolted on.Big companies already on the suitePer-action / per-conversation / per-credit (metered) on top of per-user
Developer frameworksCode libraries for building agents.Engineering teams shipping productOpen-source (free) + paid cloud/observability on top
No-code / low-code buildersVisual workflow tools that now ship with native AI agent nodes.Ops, growth, IT, citizen developersPer-workflow-execution / per-task / per-credit
Autonomous / browser agentsComputer-use agents that drive a real browser or desktop.Individuals, devs, specific verticalsPer-month subscription + usage overage
Open-source personal-agent harnessesAn 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 hackersOSS (free, BYO model) + optional managed cloud
Coding-agent IDEsAI agent that lives inside (or replaces) an editor.Software engineersPer seat (tiered) + usage overage
Voice agent platformsConversational AI that picks up the phone: low-latency speech-to-speech with function calling, telephony in/out, and a turn-taking model.
Agent observability & evalsTracing, evaluation, prompt management, and gateway logic for production agents.
EU AI Act compliance & governanceTools 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.

UnitWhat you're really paying forVolatilityWhere it shows up
Token (in + out)The LLM call itself, plus sometimes embedding/tool tokensHigh: long context, retries, thinking tokensLangSmith, AWS Bedrock, Google Gemini API
ActionA "thing the agent did": definition varies (one LLM call? one tool call? one full plan-execute cycle?)Medium: depends on agent designSalesforce Agentforce, AWS AgentCore, Manus, Replit Agent
ConversationA full session between user and agent (regardless of internal steps)Low: easy to forecastAgentforce (legacy), Copilot Studio (legacy)
Credit / LCUVendor-defined unit, usually pegged to ~1k tokens of model work + tool overheadHigh: opaque until you read the rate cardLangSmith LCU, n8n AI credits, Make, DevRev, Lindy
Task / ZAP / executionOne full workflow run, possibly multi-stepLowZapier, Make, n8n
ACU / compute unitA bundle of compute, often tied to a specific model's costMediumDevin ACUs
Per-seat / per-monthA human user license; agent usage is a separate meter on topFlatAlmost 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.

$ reference compiled 2026-06-17 · v3.1 · verified against vendor pricing pages and independent reviewers