How we tested
- Each test sent the same request three ways: with Deft, with the best free skill we found for this job, and with no skill at all.
- The judges were AI agents, not people. They saw the three results as A, B and C with every skill name removed, and scored five qualities out of 10 each, 50 in all.
- The quality scores at the top are averaged over the tests and rounded.
- How we picked the free skill: the most-starred free planning skills in our study (about 304,000 GitHub stars) and the study's pick to test against, run with the whole set readable.
- The businesses are made up. Each result is that version’s own work, unedited, from the request shown.
The work
A plan for a new app
The requestI manage 14 rental flats in Lisbon for three different landlords, and I want to make an app for it. Tenants would log in and report a problem, like a leaking tap or a broken boiler, with a photo. Each landlord logs in and sees only their own flats and the problems there, approves repairs and sees what they cost. Tenants must never see other tenants' stuff, and the landlords must not see each other's flats. I'd charge the landlords a monthly plan, maybe €12 a month per building. And I'd love an AI thing that reads the tenant's message and photo and says how urgent it is and which tradesperson to call. I'm not technical at all, I've only played with Lovable a bit. Where do I start?
The blind judge’s reasons
Deft covered every major risk in a compact plan the owner can build from, but never named EU privacy rules and assumed each building has one landlord. The free skill was the most exhaustive engineering plan but built in layers and was unreadable for this owner. The no-skill version gave the friendliest advice but wrote no plan file and told him to start building at once.
The judge’s scores, out of 10 each | Quality | Deft | Free skill | No skill |
|---|
| Finds what matters | 8 | 8 | 6 |
|---|
| A plan you can build from | 9 | 8 | 4 |
|---|
| Safe and honest | 8 | 7 | 5 |
|---|
| Fits the owner | 8 | 5 | 8 |
|---|
| The reply | 8 | 5 | 5 |
|---|
| Total | 41 | 33 | 28 |
|---|
A launch plan
The requestWe're about to launch Kettle Hill Coffee, my coffee subscription site, to real customers on 7 November at a market (I've printed 500 flyers with a QR code). Friends tested it in Stripe test mode and it all worked. What should I plan before we launch to real customers? I don't want a disaster on day one.
The blind judge’s reasons
Deft covered almost every real risk in two readable files, with dated steps that each have a clear "done when"; its one real gap was no privacy notice or terms. The free skill had the most exact engineering but no per-visitor quiz limit, no monitoring and a plan the owner can't read. The no-skill version had sharp findings but wrote no plan file, so everything sat in a 2,000-word reply.
The judge’s scores, out of 10 each | Quality | Deft | Free skill | No skill |
|---|
| Finds what matters | 9 | 8 | 8 |
|---|
| A plan you can build from | 9 | 9 | 5 |
|---|
| Safe and honest | 9 | 8 | 7 |
|---|
| Fits the owner | 8 | 6 | 7 |
|---|
| The reply | 8 | 7 | 5 |
|---|
| Total | 43 | 38 | 32 |
|---|
Adding team accounts to an app
The requestI run a flower shop in Ghent and built Petalboard with Lovable to handle our orders, deliveries and the flower stock. We use it every single day. Two other florists want to use it too, and my staff want their own logins instead of sharing mine. So I want to add team accounts (each shop with its own staff) and paid plans, something like €19 a month per shop. How do I do this without breaking anything? I'm not a developer.
The blind judge’s reasons
Deft was the only result that covers the whole request, from who can see what to payments and backups, with a short reply that asks for a yes; it missed tests that pin today's prices. The free skill was the most buildable plan by far, but left the stock race in the screens, and its reply and documents were too much for the owner. The no-skill version read the code sharply but wrote no plan file and would change the live shop directly.
The judge’s scores, out of 10 each | Quality | Deft | Free skill | No skill |
|---|
| Finds what matters | 8 | 8 | 7 |
|---|
| A plan you can build from | 8 | 9 | 6 |
|---|
| Safe and honest | 9 | 8 | 5 |
|---|
| Fits the owner | 6 | 7 | 8 |
|---|
| The reply | 8 | 6 | 5 |
|---|
| Total | 39 | 38 | 31 |
|---|