Blog · Published · Updated
AI App, AI Agent or Automation: How to Choose
Short answer
An AI app makes sense when a person needs a new place to work and approves the result themselves. An automation carries out steps in the background that you can describe before the start. An AI agent chooses the next steps based on the content of the case, within the rules. At Syntalith each form starts from €3,000 net, and delivery takes 2 to 16 weeks.
How we workDetails ↓
- 01Call30 minutes with an engineer, free
- 02PrototypeA working system on your process and data, free
- 03ImplementationTarget and fixed price in the contract; 50% at signing
- 04Acceptance50% once the agreed target is met
- net per agent for one process, build paid once
- from €3,000
- from signing the contract to acceptance
- 6–16 weeks
- free prototype on your cases
- €0
We belong to the OpenAI and Anthropic partner programmes and learn through their official courses. What this means ↓
Author: Syntalith · 4 min read
Statistics
AI rollouts: statistics 2026
2–16 wk.
is how long a rollout takes at Syntalith: automation 2–6 weeks, an app 4–10, an agent 6–16. Each form starts from €3,000 net, and an agent costs up to €17,500.
40%
According to Gartner (2025), over 40% of agentic AI projects will be cancelled by the end of 2027 because of costs, unclear value or weak risk controls.
130
According to Gartner (2025), of the thousands of agentic AI vendors, only about 130 offer real agentic solutions.
40%
According to Menlo Ventures (2025), Anthropic holds 40% of large companies' spending on language model APIs and OpenAI 27%, based on a survey of about 500 decision-makers in the US.
Source: Menlo Ventures, State of Generative AI in the Enterprise (2025) ↗
90%
According to the European Commission (2026), 90% of companies using AI say that people take the decisions, and 2% of decisions are fully automated.
Source: European Commission, Flash Eurobarometer 578 (2026) ↗
Figures checked: 22 October 2026
AI app, AI agent and automation in one table
One product can have a panel, a queue and a language model. You assess each part by who chooses the next step and who approves its effect. The name of the model and a chat window settle nothing here.
| Decision axis | AI app | AI automation | AI agent |
|---|---|---|---|
| Main user | a person working in a panel | a process triggered by an event | a system that runs the case, a person approves the risk |
| Control | the user chooses the action | code and rules lead the path | the model chooses some of the steps the rules allow |
| State | records and the user's history | the status of the run | case memory, steps, decisions, tools |
| Permissions | the role of the signed-in person | a narrow technical account | a separate scope for each tool |
| Delivery at Syntalith | 4–10 weeks | 2–6 weeks | 6–16 weeks |
One problem, three solutions
A company prepares quotes from CRM data, the price list and product documentation. An app gives the salesperson a panel: they choose the client, check the data, generate a proposal, correct it and approve it. An automation starts when a deal moves to a new stage in the CRM, assembles the document from a template and creates a task for review. An agent analyses an unusual enquiry, picks a product variant, checks for gaps, searches the approved materials and prepares a draft with questions for the client.
- Should an employee run the case in a new interface? Then the core is an app
- Should the system run with no user? Then an automation or an agent
- Can every next step be described before the start? Choose an automation
- Should the model pick tools depending on the content? An agent only in that part
- Does the action have a financial or legal effect, or change access? Stop it for approval
When the content of the case changes the next steps
| Situation | Fixed flow | Agent with a limited choice |
|---|---|---|
| A complete order form | field validation and saving by the rules | usually adds nothing |
| An email with an ambiguous enquiry | a label and a queue | choosing a source, checking gaps, a question |
| A complaint with a refund request | documents for the account owner | gathering context; a person decides |
| An integration failure | a retry and an error queue | also needs a retry limit and a manual path |
Anthropic describes this difference as a workflow with set steps against an agent that chooses the next ones itself. According to Gartner (2025), over 40% of agentic AI projects will be cancelled by the end of 2027, most often because of costs and unclear value. An agent where an automation would do is the shortest route into that statistic.
AI app, agent or automation: cost, delivery and maintenance
Process review
30 min, written plan within 2 working days
€0
An engineer counts the cost of the manual work and names what to do first.
Test on your process
5 working days
€0
The system works on 20 to 100 cases the team has already handled.
Audit
5 working days
from €600
Process map, architecture and a fixed quote. If you order the build, we credit it toward the rollout.
ChatGPT or Claude for the team
2 weeks
€1,150–4,600
Accounts, rules for working with data and training. Up to 10, up to 40 and above 40 people.
AI automation
2–6 weeks
from €3,000
A process with fixed steps: mailbox, documents, CRM, reports.
AI app
4–10 weeks
€3,000–7,000
A tool for the team, for example document search with the source.
AI agent
6–16 weeks
€3,000–17,500
Does routine steps by itself, hands exceptions to a person, records everything.
An app needs upkeep of the interface, the roles and the quality of answers. An automation needs watching over integrations, queues and retries. An agent also needs tests of its decisions, control of the cost of a single case and a check of its behaviour after a model change. Every solution has a named person who approves actions with risk and decides on process changes. In the correspondence app for foap.com, routine messages get a ready draft, and matters of money and contracts wait for a person.
How to decide
An employee needs a new place to work and approves the result
AI app
4–10 weeks
Fixed steps, triggered by an event, no user
AI automation
2–6 weeks
The next step depends on the content of the case
an AI agent in a narrow part of the process
6–16 weeks
A new screen for an employee means an app. A set process without one means an automation. An agent is needed only where the next step depends on the content of the case, and the rest stays ordinary code that is easier to test. A full overview of the forms is on the services page, and the free process review helps split the work into the right mechanisms before the quote.
