Skip to content

Chapter 7 — Toolkit Agent 2026

🧰

"40+ tools, 6 layers. Pick the layer first, the tool second."

How to use this chapter

  1. Read overview tables — know the landscape
  2. Quick picker — fast tool choice for use case
  3. Re-check every 3 months — Anthropic + OpenAI release fast

01 Layer 1: Coding Agents

ToolPrice (May 2026)StrengthBest for
Claude Code$20 Pro / $100-200 Max / APITerminal-native, sub-agent, 30h autonomous, MCP-nativeMulti-file refactor, large codebase
Cursor$20/month Pro (credit-based)Best autocomplete, polished IDEDaily inline coding
Windsurf (Cascade)$15/month Pro (cheapest top 3)"Flows" persistent contextValue-conscious
Devin 2.0$20 Core / $500 TeamCloud-based, parallel autonomous PRLong tickets, enterprise
OpenAI CodexBundle ChatGPT Plus/Pro2M+ WAU Mar 2026, cloud sandboxChatGPT users, cloud async
GitHub Copilot Workspace$10-39/user/monthGitHub-nativeExisting GH enterprises
Replit Agent$25/month CoreBrowser-based, full-stack scaffoldNon-dev MVP
AiderFree + your API key45.3K stars, terminal, git-awareOSS purists
ClineFree + your API key62.3K stars, "trusted by 8M+", VS CodeVisual diff approval

Quick picker

Use caseTop pickBackup
Multi-file refactor large codebaseClaude CodeDevin
Daily inline coding flowCursorWindsurf
Budget-tightWindsurf or AiderCline
Cloud async (long task)Devin or OpenAI Codex
Non-dev build MVPReplit AgentLovable
OSS / privacy concernAider or Cline

02 Layer 2: General-Purpose Agents

ToolPriceStatus (May 2026)
Manus AI$199/month ProBlocked by Chinese NDRC Apr 2026 — Meta acquisition $2-3B unresolved
Anthropic Computer UseAPI Sonnet 4.6 $3/$15Beta computer-use-2025-11-24; 72.5% OSWorld-Verified
OpenAI ChatGPT Workspace AgentsBundle enterpriseApr 2026 — Operator absorbed
Project MarinerShutdown May 4, 2026 — absorbed into Gemini Agent
Claude Code SDKAPI-pricedOrchestrator-via-Task tool, MCP-native

Quick picker

Use caseTop pick
Automate legacy app no-APIAnthropic Computer Use
ChatGPT-native workflowWorkspace Agents
Build custom agentClaude Code SDK

03 Layer 3: Workflow / Automation

ToolPriceNumbers (May 2026)Best for
n8nFree / $20+ cloud / $13.3K avg ARR enterprise$40M ARR, $180M Series B, $2.5B valuation, 230K users, 3K+ enterpriseFull-stack workflow, AI nodes
Make$9-29/month1,500+ appsVisual scenarios
Zapier Central/AgentsVariableEstablishedExisting Zapier users
Smax.ai (Vietnam)$25-89/monthYody +15-20% close, Let's Sushi +300% onlineVN Messenger/Zalo bot
LindyFree / $49.99 Pro / $299.99 Business2,300+ appsNo-code agent
Sema4.aiEnterprise$55.5M raised, Emerson + KochFinance/HR enterprise

Quick picker

Use caseTop pick
Full-stack workflow + AIn8n
Vietnamese Messenger/ZaloSmax.ai
No-code agent for allLindy
Visual scenarios non-techMake
Enterprise SAFESema4.ai

04 Layer 4: Multi-Agent Frameworks

FrameworkStatus May 2026Best for
LangGraphProduction powerhouse — Klarna, Uber, LinkedIn. Surpassed CrewAI stars early 2026.Mission-critical production
CrewAIRole-based DSL, 20 lines to start. 60% Fortune 500, 44K+ starsPrototype, agency
AutoGen 0.4+Microsoft maintenance modeDon't start new
OpenAI Swarm → Agents SDKLightweight handoff-basedOpenAI-native
Anthropic / Claude Code SDKOrchestrator-worker via Task tool, MCP-nativeClaude-first stack
A2A protocolOpen standard Apr 2025, donated Linux Foundation Jun 2025, 150+ supportersCross-vendor agents

Quick picker

Use caseTop pick
Production scaleLangGraph
Prototype fastCrewAI
Claude-nativeClaude Code SDK
OpenAI-nativeOpenAI Agents SDK
Cross-vendor agentsA2A

05 Layer 5: Infrastructure

ToolDifferentiatorCustomers
E2BmicroVM, ~150ms cold start, 500M+ sandboxes processed, 88% Fortune 100 signedPerplexity, Hugging Face, Groq, Manus
BrowserbaseHeadless browser as a service, 1K+ paying customers in 16 months, $40M Series B at $300M valuation
DaytonaDocker-based, <90ms sandbox creation, $24M Series A Feb 2026
Modal / Vercel SandboxServerless compute for agents
MCP servers ecosystem5,800+ official, 17,468 cross-registry, 97M monthly SDK downloadsAnthropic, OpenAI, Google all adopted

Quick picker

Use caseTop pick
Code execution sandboxE2B
Browser automationBrowserbase
Cheap Docker sandboxDaytona
MCP serverAnthropic SDK + Smithery

06 Layer 6: Voice Agents

ToolLatencyCost/minStrength
ElevenLabs Conversational AI<100ms$0.08-0.2411K+ voices, 70+ languages
Vapi~150ms$0.05 orch + providerProvider-agnostic, 62M monthly calls, 99.99% SLA
Retell AI~620ms$0.07+ flatHIPAA included, no-code + SDK
Bland~400msVariablePhone-first

Quick picker

Use caseTop pick
Multi-language brandElevenLabs
Custom stack + scaleVapi
Healthcare HIPAARetell
Outbound campaignsBland

07 5 ideal stacks by persona

Persona 1: Solo dev (budget < $100/month)

Coding: Cursor Pro ($20) or Claude Code Pro ($20)
LLM API: Claude Haiku 4.5 + Sonnet 4.6 (~$30/month usage)
Sandbox: E2B free tier
Deploy: Vercel free + Supabase free
─────────────
Total: ~$60-80/month

Persona 2: Indie agent builder

Coding: Claude Code Max ($100)
Workflow: n8n cloud Starter ($20)
Infra: E2B ($20) + Browserbase free tier
Voice: Vapi ($30 usage)
─────────────
Total: ~$170/month

Persona 3: Agency (serving SMEs)

Coding: Claude Code Pro ($20)
Workflow: Smax.ai Pro ($89) + n8n Pro ($50)
LLM: Claude API ($50-200 usage)
Voice: Vapi ($50-200)
CRM: Pancake ($30)
─────────────
Total: ~$280-580/month (charge $5-15K project)

Persona 4: Enterprise team (10+ devs)

Coding: Claude Code Max × 10 devs ($1,000-2,000)
Workflow: n8n Enterprise ($1K+)
Multi-agent: LangGraph (LangSmith $250+)
Voice: Vapi enterprise ($500+)
Eval: Braintrust ($500)
─────────────
Total: ~$3K-5K/month

Persona 5: Operator non-dev

Workflow: Lindy Pro ($49.99) or Make ($29)
LLM: ChatGPT Plus ($20)
Voice: Retell no-code tier
─────────────
Total: ~$70-100/month

08 Selection framework

Decision tree

Q1: What's the use case?

  • Solo coding → Layer 1 (Claude Code / Cursor)
  • Automate workflow → Layer 3 (n8n / Smax.ai)
  • Custom agent product → Layer 4-5 (LangGraph + E2B)
  • Voice → Layer 6 (Vapi / ElevenLabs)

Q2: Budget?

  • < $100/month → free tier + 1 Pro tool
  • $100-500/month → multi-layer indie stack
  • $500/month → enterprise tier or team

Q3: Regional context needed?

  • Yes (target VN customer) → Smax.ai + Zalo OA + Pancake mandatory
  • No (global) → standard global stack

Q4: Skill level?

  • Non-dev → Lindy + Make + Replit Agent
  • Junior dev → Cursor + n8n + ChatGPT
  • Senior dev → Claude Code + LangGraph + custom MCP

09 Continue reading

Final word

"Don't memorize tools. Memorize layer roles.Coding agent = augment human dev.Computer Use = automate legacy.Workflow = orchestrate cross-app.Multi-agent = parallel specialists.MCP = standard interconnect.Voice = phone/call interface.Tools rotate. Layers stay."