Short answer: an AI agent is a system where the LLM plans its own steps, picks its own tools (APIs, databases, email) and keeps going until the job is done. A chatbot answers questions, and automation runs steps a person fixed in advance. There are two costs: a one-off build (quoted per scope) and monthly API usage, which for 1,000 tasks a month runs about THB 110–2,200 depending on the model. For most Thai SMEs, automation or a chatbot is the better buy. Move to an agent only when the input is messy and the steps cannot be fixed in advance.
What an AI agent is: a definition you can decide with
Anthropic draws the line in Building Effective Agents (19 Dec 2024): a workflow is a system where LLMs and tools are orchestrated through code paths written in advance, and an agent is a system where the LLM directs its own process and tool use. So one question settles it: who decides the steps? A person: automation. The AI: an agent. Agentic AI names the same approach, so a buyer can treat the two as one. Google Ads data (via DataForSEO) shows Thais searched for ai agent คือ 1,000 times in Sep 2025, peaking at 4,400 in Jun 2026 and sitting at 2,900 in Aug 2026. Rising interest does not mean every business needs one.
Chatbot, automation, AI agent: what is the difference?
All three can use an LLM. What differs is the job and who controls the steps. For the chatbot side, see AI chatbot pricing and RAG explained.
| Aspect | Chatbot | Automation | AI agent |
|---|---|---|---|
| Job | Answers questions | Runs fixed steps | Plans its own steps |
| Who decides the steps | No steps: question in, answer out | A person, in advance | The LLM, as it goes |
| Input | Customer messages | Same format every time | Many formats, messy |
| Cost per task | Low (one LLM call) | Lowest (may need no LLM) | Highest (several LLM calls) |
| Example | Price and shipping answers on LINE OA | Web form → Sheets → LINE alert | Read order emails, check stock, draft a quote |
Which should a Thai business choose?
Anthropic's own advice is to use the simplest solution first, because agents trade latency and cost for performance. That fits most Thai SMEs, whose time goes on repeated questions and moving data between systems. Neither needs an agent:
- •Customers ask the same things (price, shipping, opening hours) → a chatbot on LINE OA or your website
- •The steps are the same every time (order in → record → notify) → automation, for example n8n with LINE OA
- •Input arrives in many formats and each case needs different handling (customer emails, PDFs, photos of purchase orders) → an AI agent
💡 Simple rule: you can write every step on paper → automation · customers only ask questions → chatbot · you cannot write the steps in advance because no two inputs look alike → agent
What an AI agent costs: two separate bills
Bill one: the one-off build. It depends on scope: how many systems to connect, how risky the task is, and where a person has to approve. chercode quotes per project after a scoping call, so there is no standard figure. Bill two: monthly API usage, paid per token. Official prices per 1M tokens are below (baht at roughly THB 36.5 per USD). For full plans see Claude pricing, or call several models with one key through OpenRouter.
| Model | Input | Output |
|---|---|---|
| Claude Haiku 5.5 | $0.10 (≈฿3.65) | $0.50 (≈฿18) |
| Gemini 3.8 Flash | $0.75 (≈฿27) | $3.75 (≈฿137) |
| Claude Sonnet 5.5 / GPT-6.1 Sol | $2 (≈฿73) | $10 (≈฿365) |
⚠️ The Haiku 5.5 price applies to prompts up to 100,000 tokens ($0.50 / $2.50 above that) · the Gemini 3.8 Flash price holds until 31 Dec 2026, then rises to $1.50 / $7.50 from 1 Jan 2027 · verify on platform.claude.com/docs/en/about-claude/pricing, ai.google.dev/gemini-api/docs/pricing and developers.openai.com/api/docs/pricing · updated 12 Oct 2026
Real numbers: three example Thai businesses
All three are hypothetical. The figures are estimates from the stated assumptions and exclude the build cost:
- •Say an online clothing shop gets 3,000 customer messages a month on LINE OA → chatbot. At 3,000 input + 300 output tokens per message on Haiku 5.5: (9M × $0.10) + (0.9M × $0.50) = $1.35, about THB 50 a month. Replies sent through the Messaging API do not count against the quota, so the Free plan (300 messages) is enough unless you broadcast. Then Basic is THB 1,280 a month for 15,000 messages and Pro THB 1,780 for 35,000, before 7% VAT
- •Say a contractor gets 500 quote-request forms a month that must go into Sheets and alert the team → automation, no LLM needed. n8n Cloud Starter is €20 a month (about THB 800, billed annually, 2,500 executions) and Make Core starts at $9 (about THB 330, 10,000 credits). Self-hosted n8n Community Edition is free for internal use, so you pay only for the server
- •Say an importer receives 1,000 order emails a month as PDFs, photos and plain text, each needing a stock check and a draft quote → agent. At 20,000 input + 2,000 output tokens per task across all LLM calls: 20M + 2M tokens a month. Sonnet 5.5: $40 + $20 = $60, about THB 2,200 · Gemini 3.8 Flash: $15 + $7.50 = $22.50, about THB 820 · Haiku 5.5: $2 + $1 = $3, about THB 110
⚠️ Converted at roughly THB 36.5 per USD and THB 40 per EUR · n8n Community Edition is under the Sustainable Use License, which is not an open-source licence · push and broadcast messages do count against the LINE OA quota · verify on n8n.io/pricing, make.com/en/pricing, developers.line.biz and lineforbusiness.com/th · updated 12 Oct 2026
Limits to know: agents still fail about one time in three
The Stanford AI Index 2026 reports that agent task success on OSWorld (a benchmark of real computer tasks) rose from 12% to about 66%. A real jump, yet agents still fail roughly one attempt in three on a structured benchmark, and real company work is messier.
- •Keep a person approving. For anything touching money, customer data or outbound messages, the agent drafts and a person confirms. The layer that enforces this is the agent harness
- •Projects do get cancelled. In June 2025 Gartner forecast (from a January 2025 poll, reported by IT Brief) that over 40% of agentic AI projects will be cancelled by the end of 2027 because of rising cost, unclear business value and weak risk controls. That is a 2025 forecast, not a 2026 measurement
- •Be careful building on a vendor's visual builder. OpenAI is shutting down Agent Builder on 30 Nov 2026, so anyone who built on it has to move
Verdict: does your business need an AI agent?
Most do not, yet. If you can write the steps on paper, use automation. If customers only ask questions, use a chatbot. Both are cheaper, faster and easier to test. An agent earns its cost when the input is messy and each case needs a different decision. When you get there, start with one task, try a cheap model first, and keep a person approving anything that touches money or customers.
🎯 Not sure whether your task should be a chatbot, automation or an agent? See the AI Agent Development service or send over the details of your task. If plain automation is enough, you will be told so.
Arm - chercode
Full-Stack Developer & Founder
Software developer with 5+ years of experience in Web Development, AI Integration, and Automation. Specializing in Next.js, React, n8n, and LLM Integration. Founder of chercode, building systems for Thai businesses.
Portfolio
