The AI Agent That Assumes: Why Automation Needs to Ask, Not Guess
AI agents are good at filling in gaps quietly, which is exactly the problem. Here's why business automation needs to be built to ask rather than assume, with real examples.
API & systems integration
Get your systems talking to each other
Most businesses run on a handful of tools that don't share data, so people spend their days copying information between them. I connect those systems so the information flows on its own, and that manual work goes away.
01The problem
Your accounting package, CRM, website and supplier systems each hold part of the picture, and none of them talk. So someone re-enters the order into the system, copies the customer into the CRM, updates a spreadsheet and sends the confirmation by hand. It's slow, it's easy to get wrong, and it scales badly. Integration replaces all of that with information that moves by itself.
02The outcome
What changes when your systems connect
The point isn't the technology, it's what it does for how the business runs.
Information flows on its own
An order, enquiry or update entered in one place shows up everywhere it's needed, without anyone copying it across.
One version of the truth
Your systems agree with each other, so the numbers you report and the decisions you make are based on data you can trust.
Less manual admin
The repetitive re-keying, exporting and reconciling that eats your team's time simply stops being necessary.
Fewer mistakes
Every manual hand-off is a chance to mistype or forget something. Automating the flow removes that whole class of errors.
03Examples
The kinds of things I connect
A few common examples. If the systems you use aren't here, the approach is the same.
Payments
Stripe and other payment providers connected to your systems so sales, subscriptions and refunds reconcile automatically.
CRM & sales tools
Keep your CRM in step with your website, accounts and operational systems, instead of updating each by hand.
Email & SMS
Trigger the right message to the right person at the right moment, driven by what's actually happening in your systems.
AI & third-party APIs
Bring external services and AI APIs into your workflow where they genuinely help, wired in reliably.
Webhooks & data sync
Real-time updates between platforms so a change in one is reflected in the others within seconds, not overnight.
Import & export
Move data cleanly between systems, including the one-off migrations and the ongoing scheduled transfers.
04How it runs
How I approach an integration
- 01
Understand the systems
I learn what you're running, what each system is responsible for and where information currently gets stuck or re-entered.
- 02
Map the data and workflows
We map how information should flow between them, and what a change in one system should do to the others.
- 03
Design the integration
I design how the systems will connect, choosing the right method for each, from proper APIs to webhooks to scheduled syncs.
- 04
Build it
I build the integration to be reliable and clear, so it can be understood and maintained rather than being a black box.
- 05
Test the failure cases
The important work is what happens when something goes wrong: a system is down, a message is duplicated, data is malformed. I test and handle those.
- 06
Deploy and monitor
It goes live with monitoring so problems are visible, with retries and reconciliation so the occasional hiccup doesn't lose data or need a person.
05FAQs
Integration questions
How do you connect two systems that weren't built to talk?
It depends on what each system offers. Many have proper APIs or webhooks I can use. Where a system is more closed, there are usually other routes, from scheduled data exports and imports to database-level integration. Part of the job is finding the most reliable method for each case.
What if a connected system goes down or sends bad data?
That's exactly what a well-built integration plans for. I design in retries, sensible error handling and reconciliation, so a temporary outage or a malformed message doesn't lose data or leave your systems out of step. When something needs a human, it flags it rather than failing silently.
Will this reduce our manual admin?
That's usually the whole point. If your team spends time copying information between systems, re-keying the same details or exporting and reformatting data, connecting those systems removes that work and the errors that come with it.
Do you work with our existing tools?
Wherever possible, yes. The aim is to make the tools you already pay for work together, not to replace them. If a tool genuinely can't do what you need, I'll tell you, but that's the exception.
Is this a one-off project or ongoing?
Either. Some integrations are build-and-leave; others need ongoing support as the connected systems change. I'm happy to do both, and I'll keep an integration running smoothly if you'd like me to.
What if I need a bigger system, not just a connection?
Sometimes the right answer is a proper bespoke system rather than more integrations bolted on. If that's the case I'll say so. Have a look at custom software, or just tell me the problem and I'll point you the right way.
Insights
Related reading
You Inherited a System Nobody Can Explain. Now What?
When you inherit software with no documentation and no developer to ask, the temptation is to rewrite it fast. Here's a safer way to work out what it actually does first.
Your AI Coding Assistant Doesn't Remember Last Week's Decisions
AI coding agents don't carry memory of past decisions between sessions the way a human developer does. That gap shows up as inconsistency in your codebase, and it matters more than most people realise.
What needs connecting?
Tell me which systems you're running and where the manual work is. I'll tell you what can be connected and what it would take.
Or email me at josh@thackr.co.uk