Compare
Custom Scripts
hand-written code, often in Python, Playwright, or AutoHotkey, that automates a specific task by driving software or files directly
Who they are and what they are good at
All three are sound tools in the right hands.
- Python with PyAutoGUI. Moves the mouse and presses keys on Windows, macOS, and Linux. It can find an image on screen but does not read text, and it sees the primary monitor only. On Windows, pywinauto reaches controls by name through UI Automation.
- Playwright. Open-source browser automation for Chromium, WebKit, and Firefox, with a click recorder and a Trace Viewer that captures screenshots and DOM snapshots per action.
- AutoHotkey. A free Windows scripting language. Its ImageSearch finds an image only at exactly the size it was captured.
What they share is an owner: the person who wrote the script is the only one who can change it.
Where scripts stop for messy back-office work
A script covers exactly what its author anticipated, on the screens that existed the day it was written.
-
Second factors
A fuel card portal with a code prompt needs a person. Playwright's saved sign-in file expires, and its own docs warn the file can be used to impersonate you.
-
Screens that move
Selectors, coordinates, and image matches fail when a page or resolution changes, and the fix waits for the author.
-
Judgment
A card swiped for a truck that sat in the yard all day ends up in a message box, if anywhere.
-
Evidence
A script records only the log lines its author chose to write.
-
Ownership
When the author leaves, so does the knowledge of what it touches.
LaunchAI runs your transform as a code hook, works the portal in the equipment manager's session, applies exact rules, and sends exceptions to a review step. More on Product →.
Side by side
| Python with PyAutoGUI | Playwright | AutoHotkey | LaunchAI | |
|---|---|---|---|---|
| Who builds | A developer | A developer | A power user | The person who does the work |
| Who can change it | Whoever reads the code | Same | Same | The process owner, by editing a table or re-showing a step |
| Reach | Any screen, by coordinates or images | Browsers only | Windows apps | Any app a person can use, plus APIs |
| Reads text on screen | No | Page text from the DOM | No | Yes |
| When a screen changes | Match fails | Selector fails | Match fails | Re-plans; stops for a person if unsure |
| Sign-in and second factor | Stored credentials | Saved state file | Stored credentials | The user's session; codes asked of a person |
| Exceptions | Whatever the author anticipated | Same | Same | A review step with the source beside the record |
| Evidence | Author's log lines | Browser traces | Author's log lines | Screenshots of every action |
| Speed per step | Fast | Fast | Fast | Human pace, verified |
When to choose a script
API calls between systems your team controls.
Browser regression tests for your own site.
One-time cleanups and personal hotkeys.
When to choose LaunchAI
- Fuel card reconciliation: portal download with a code prompt, card-to-unit mapping, tank capacity and telematics checks, posting in Spectrum, and a statement that must tie to the cent.
- The process owner, not a developer, should keep it running.
- You need an answer when an auditor asks what ran last Tuesday.
Keep the script, retire the fragile parts
Keep the transform. It becomes a code hook on one step. The agent handles the login, the screens, the exceptions, and the evidence around it. Move hard-coded lookups like card numbers and tank sizes into tables, delete saved sign-in files from shared drives, and switch old scripts off one at a time.
BeforeAfter
FAQ
Why did our AutoHotkey script stop after the new monitors arrived?
Image matching is literal. A new resolution or scaling setting changes every pixel it was looking for.
Can we keep a Playwright sign-in file on the shared drive?
Treat it like a password. Anyone who can read it can sign in as that user.
When should a script be retired instead of wrapped?
When the process owner cannot read it, it breaks more than once a quarter, or nobody can say what it did last Tuesday.
Can LaunchAI run the Python we already have?
Yes, as a code hook in a restricted interpreter. Confirm how your imports fit during a demo.
No prompts. No babysitting. No Integrations.