❯ Question and answer
When is it not worth using AI in an automation?
It is not worth it when the rule is fixed and the input already arrives structured: copying a form lead into the CRM, sending a confirmation email, issuing an invoice. A direct flow without AI is cheaper, faster and cannot misread anything. AI earns its cost when you need to understand free text or decide case by case.
Updated on
What are the numbers, with sources?
- Anthropic recommends finding the simplest solution possible and only increasing complexity when needed; agentic systems trade latency and cost for better task performance. (Source: Anthropic, Building effective agents)
- In October 2026, Claude Sonnet 5.5 costs US$2 per million input tokens and US$10 per million output tokens: every message processed by AI has a cost, a fixed flow does not. (Source: Claude API pricing)
- OpenAI’s guide reserves agents for 3 cases: complex decision-making, rules that are hard to maintain and heavy reliance on unstructured data. (Source: OpenAI, A practical guide to building agents)
Which tasks work better without AI?
The ones with the same input and the same output every time:
- Moving a lead from the website form into the CRM.
- Sending confirmation, reminder or billing emails with data already in the system.
- Copying data between a spreadsheet, the CRM and finance.
- Alerting the team on WhatsApp or email when a field changes.
When does AI start paying off?
When the input is free text or the decision changes case by case: understanding what a customer asked for in a message, classifying an email, summarizing a conversation, answering varied questions on WhatsApp. A fixed rule would break all the time there.
Can I mix fixed flows and AI?
Yes, and it is usually the best design. AI handles only the step that needs interpretation (classifying the message, extracting the data) and the rest of the path follows fixed rules, which are cheaper and predictable.
Common mistakes
- Using AI to move data that already arrives structured, such as a form with fixed fields.
- Letting AI make decisions the company already has as written rules, such as discounts or deadlines.
- Not measuring the cost per message before scaling an AI flow to thousands of contacts.
- Dropping AI altogether because the first test was a loose prompt with no data or integrations.
How does Notkode decide whether AI goes in?
Automation scope at Notkode comes in 3 tiers: simple (1 direct flow, no AI), intermediate (2 to 3 flows with light AI for classification and automatic replies) and advanced (an AI agent with conditional logic and multiple integrations). AI only goes in at the tier where it truly replaces manual work.
Frequently asked questions
Is an automation without AI still worth building?
Yes. A large share of a company’s manual work disappears with fixed flows, with no language model involved.
How much does an AI call cost?
It depends on the model and the length of the text. In October 2026, Claude Sonnet 5.5 charges US$2 per million input tokens and US$10 per million output tokens.
Can I start without AI and add it later?
Yes. A well-built fixed flow becomes the base: when a step needs to interpret text, AI goes into that step only.
How do I know if my case needs AI?
Ask whether the input is always the same. If it is, a fixed rule does the job. If it is free text or the decision changes case by case, AI tends to pay off.
Last verified on