The short version
Software is faster to start and cheaper in the first months. A custom agent costs more upfront and fits your processes exactly.
The deciding question is where the agent must act. If it only answers and drafts, a tool is often enough. If it must update your CRM, accounting or booking system with the right fields, custom work usually wins.
Side-by-side comparison
The table compares the two options on the points that usually decide the choice.
Ready-made agent software vs custom agent| Point | Agent software | Custom agent |
|---|
| Start | Days, by your own team | Scoping, build and testing with a developer |
|---|
| Cost model | Subscription per seat, channel or usage | Build price plus hosting and model usage |
|---|
| Your systems | Only the connectors the vendor offers | Any system with an API, export or shared file |
|---|
| Zalo OA | Depends on the vendor; check support for your account | Connected through the official Zalo OA API |
|---|
| Rules and approvals | Within the tool's settings | Written to your process, including human approval steps |
|---|
| Data location | In the vendor's cloud | In accounts you own |
|---|
| Changes | Wait for the vendor's roadmap | Change it yourself or with your developer |
|---|
When ready-made software is the better choice
Pick software when your need looks like everyone else's. Many tools handle common jobs well and keep improving at no cost to you.
- You need FAQ answers and lead capture in one channel
- Your CRM or helpdesk already has an AI add-on that covers the task
- Your volume is small, so a build would not pay back
- You want to test whether customers accept an AI reply before investing more
- You have no developer and no budget for monthly support
In these cases we recommend buying software, even though it means no build project for us.
When a custom agent pays off
Custom work pays off when the value sits in your own systems and rules. A generic tool cannot see your price list, your stock or your approval chain unless someone connects them.
Vietnamese documents are another reason. E-invoices, delivery notes and contracts in local formats need testing on your real samples, and tools tuned on English data may misread them.
- The agent must create or update records in your CRM, ERP, accounting or booking software
- Steps need human approval, for example a quote above a set amount
- Customer or patient data must stay in accounts you control
- You need English and Vietnamese with different answers per market
- Subscription fees per seat grow faster than the work the agent saves
Costs that are easy to miss
Both options have costs outside the first invoice. Put them in one table before you decide.
Software adds seats, channels and usage tiers as you grow. Custom agents add hosting, model usage and the time to keep answers and rules up to date.
Switching costs matter too. Moving from one tool to another means rebuilding flows and exporting history, so ask how you get your data out before you start.
How to test the choice in two steps
Do not decide from demos. Collect a set of real cases from your inbox or documents, with personal data removed.
Run the cases through the tool's trial. Note where it fails, where staff must step in and what it would cost at your volume.
If the tool fails on cases that matter, scope a custom agent for those cases only. A mixed setup is common: software for simple replies, a custom agent for the steps inside your systems.