Journal
The journal from the studio.
Our choices and opinions, and more: practical help for launching a product without a technical background. Written plainly.
Bubble or Lovable: two ways to lock you in, and the question to ask
One interprets your application in an engine found only on its platform. The other lets you leave with code but refuses to take yours in. The locks are different, and the question separating them fits in one sentence.
Read the article
AI breaks my application with every change: why a method beats a better model
You request one fix and get three failures. The reflex is to switch tools, then models, then tools again. Those who escaped the loop describe a writing habit costing one line per request.
Read the article
Custom restaurant software: ingredient costs, purchasing and schedules, without the till
A compliant till is a legal obligation and will remain so: a tool you have built must not take payments. Everything else, ingredient costs, purchasing, hygiene plans and scheduling, is currently rented as modules from vendors, two of four of whom refuse to display a price.
Read the article
Firebase Studio is closing: what to replace it with, and what a closing tool owes you
Google closed Firebase Studio signups in June and will delete workspaces in March 2027. GitHub deprecated Spark in August. Two application-building workshops ended in one summer, with the same lesson for those building elsewhere.
Read the article
Custom software or SaaS: the five-year calculation, including per-user subscriptions
SaaS costs little in month one and a lot in year five, especially when billed per user. Custom software does the reverse. Here are both columns side by side, with French agencies' public prices and the maintenance line quotes leave out.
Read the article
Why my no-code MVP never gets finished: day 9 and the remaining 20%
The screens exist, the demonstration works and the test payment succeeds. Then the project stops and nobody can say why. An account from a no-code MVP provider, what the final fifth contains, and why it costs dearly after a launch date has been announced.
Read the article
Custom software for physiotherapists and osteopaths: session series, unpaid bills and fee sharing
Practitioners do not complain about electronic claims; they complain about prices rising without warning and second logins charged at undisclosed rates. What custom software handles, and what it leaves to approved software.
Read the article
Automated decisions and GDPR: €825 million for missing human approval
An automated decision depriving a driver of their account without human review: the Dutch authority, working with the CNIL, valued the failure at €824,990,000. Human approval is covered by an article of GDPR, and its absence has a price.
Read the article
Lovable review 2026: what $25 buys, and what you take when you leave
Pro starts at $25 per month and top-ups cost $0.30 per credit. What remains is to establish what a credit buys, what is consumed while you do nothing, and what you retrieve when you decide to leave with your application.
Read the article
Exposed customer data: a two-minute test for an AI-created application
Your application displays a login screen, so your customer data is protected. Providers repairing these applications describe the opposite across three different samples. Three checks you can perform tonight, and three questions for whoever builds it.
Read the article
Custom software for independent lawyers: cases, deadlines and fees without add-ons
One lawyer in three works alone and pays for software per user, then per module. What legal software vendors cover, what costs extra, and the layer left open by RPVA and CARPA, with prices from four vendors collected on September 2, 2026.
Read the article
My developer has stopped responding: what must be in your name
My developer has stopped responding: the question always arrives at the worst moment, and the bad news lies somewhere other than code. It comes from accounts, access and the domain name left with someone else. Here is the list to check before you need it.
Read the article
Building an application without knowing how to code in 2026: four routes
No-code, prompt generator, coding assistant, agent team: four ways to have software written without writing it yourself. They do not cost the same, and above all they do not leave the same things in your hands when you stop paying.
Read the article
Claude Fable 5.1 pricing: the reduction is in the cache, not the headline rate
Anthropic introduced Claude Fable 5.1 on September 1. The per-token rate is unchanged; rereading context falls by 75%. In practice, long building sessions that keep the same project in view for hours become the cheapest way to work.
Read the article
Application development is conducted like an orchestra
A score written before playing, sections with clear remits, a conductor holding the baton without playing an instrument: the metaphor structuring Maestro, embraced all the way to its logo.
Read the article
How long does it take to create an application in 2026?
Hours, days or months for the same idea: the difference lies less in writing code than in the exchanges around it. Timelines, route by route.
Read the article
Why agency prices have not followed the fall in AI costs
What cost $60 per million tokens in late 2021 now costs between $0.06 and $0.40 at equivalent performance, a reduction of 150 to 1,000 times. Meanwhile, a developer's average daily rate in France remains around €629. Someone is keeping that margin.
Read the article
Migrate your Excel data into your application: a guide without breakage
The application is ready; the historical file is waiting. Migration happens in four stages: clean in Excel, turn formulas into rules, import with a discrepancy report, and run both in parallel.
Read the article
AI makes experts slower: why approval must be an architecture, not a chore
The METR study measured experienced developers as 19% slower with AI, while they believed they were 20% faster. Review and verification consumed the gain. The problem is not review: it is where we put it.
Read the article
The 5 AI development traps nobody tells you about
Drift without specifications, false green lights, opaque credits, disposable code and dependence: five documented traps encountered in our own measurement campaigns, and a safeguard for each.
Read the article
Self-service skills: the supply chain nobody is watching
The github/spec-kit repository shows 130,054 stars a year after launch. The AI agent skills ecosystem is growing just as fast, and installing a skill on demand is becoming the new curl piped into a terminal.
Read the article
A prototype in an evening, a product in a week: the real boundary
AI tools make spectacular prototypes and viable products possible. Between them lie three gaps invisible on screen: written rules, tests that prove, and versions that can be restored.
Read the article
What our beta testers broke, second pass: the silence that blocked everything
After a first real user test, 21 written comments revealed a defect invisible internally: an entire step type was handled nowhere in the engine. The chain froze silently, waiting for an answer that never came.
Read the article
What happens to your application if the tool that created it disappears?
Platforms closing, prices tripling, acquisitions: the scenario deserves a rehearsal before you choose. Two assets determine what happens next: your code and your specification. The exit test comes down to three questions.
Read the article
What does an application built by agents really cost? Our bills, line by line
Unlimited AI plans are dying, and budgets are exploding at businesses without caps. We publish our real costs: $105.64 for a complete campaign of 5 judged tasks, from brief to verified product.
Read the article
Your requirements document in one hour: the awkward-questions method
A timer, four blocks: ten minutes for the problem, twenty for journeys, twenty for awkward questions and ten for the list of things to exclude. The complete step-by-step method.
Read the article
Tests first, and woe to the agent that changes them: TDD imposed on machines
TDFlow reaches 88.8% on SWE-Bench Lite when tests are supplied before code. At Maestro, tests have been written BEFORE each story since July, with one simple rule: an agent modifying a test, even a pre-existing one, is flagged.
Read the article
Accountant, physiotherapist, architect: custom software for independent professionals
Regulated professions share a problem: detailed business rules flattened by generic software. Specifications approved by the professional, now within reach of a practice, preserve those rules as they are.
Read the article
“Looks good to me”: why human approval is AI’s real safeguard
Article 14 of the European AI Act requires ‘effective’ human oversight of high-risk AI. At Maestro, nothing is agreed without ‘Looks good to me’: not a regulatory constraint added later, but the product's central mechanism from day one.
Read the article
Association software: manage members, tax receipts and general meetings without a budget
The volunteer treasurer juggles four files and an email folder. You can now direct the building of a custom tool managing members, receipts and meetings for the price of a subscription, around €20 per month.
Read the article
Which drifted, the system or the judge? We measured our jury’s strictness
An evaluation of 21 AI judges shows rankings shifting by up to 14 places depending on the benchmark. Our average score dropped six points between campaigns: we had to prove whether our engine or our jury had changed.
Read the article
No-code or AI: should you still learn Bubble in 2026?
No-code promised building without coding, at the cost of learning an editor. AI shifts the promise: nothing left to assemble yourself, requirements to direct, and standard code you own.
Read the article
Therapists and care professionals: confidentiality as the number-one requirement
On November 6, 2025, Doctolib was fined €4.6 million for abuse of a dominant position. In June 2026, health data was being used to train AI models. For a therapist, the question is no longer which software, but where their notes live.
Read the article
Custom management software for tradespeople: generic SaaS or your own tool?
Obat and similar software cover construction quotes and invoices well. The question arises when your way of working outgrows their boxes. Here is the tipping point, and what each side costs.
Read the article
Our benchmarks publish our defeats: what SWE-bench can no longer tell you
32.7% of successful SWE-bench fixes contain solution leakage, according to a recent position paper. As the reference leaderboard collapses, we publish our own duels, including BOTH victories and defeats.
Read the article
Can an AI agent team deliver your software? What we measured
Nine projects taken from idea to verified product, three judges per submission, costs published to the cent. What measurement says about an agent team's real capabilities, and what it does not.
Read the article
Why we left BMAD: an autopsy in numbers
Maestro started with BMAD-METHOD. We left it for our own engine, Partition, and measured the replacement campaign after campaign: an average of 89.5 versus 83.5 across nine tasks, with eight duels won out of nine.
Read the article
Vibe coding explained to people who will never code
Describe what you want and let AI code it: vibe coding won over developers in 2025. For a non-developer, one thing is missing: a place to record decisions.
Read the article
“Waterfall strikes back”: our response, point by point
225 points and 191 comments: the Hacker News thread accusing spec-driven development of reviving waterfall made an impression. Three precise criticisms, a quantified response, with no evasion.
Read the article
Lovable, Bolt, v0 or Maestro: which AI tool should you choose for your application?
Compare project preparation, construction, publishing and code handover. A reading of documented capabilities to help you choose for your needs, including Maestro's limitations.
Read the article
Accountants and their internal tools: what custom software unlocks
From September 1, 2026, electronic invoicing becomes mandatory through an approved platform (impots.gouv.fr). Seven in ten practices are microbusinesses: the ideal moment to replace Excel workarounds with internal tools of their own.
Read the article
Rebuilding business software: when and how should you proceed?
When should you rebuild a business application? Signals to examine, stages, data migration and budget: a method for deciding and preparing the transition.
Read the article
Our harshest tester is an agent: five temperaments, from rushed to destructive
Since July, every new engine version has faced an agent playing a user with five temperaments, from rushed to destructive, capped at $15: five real bugs caught on day one.
Read the article
Replace Excel with an application: choose the right solution and migrate
Should you improve your spreadsheet, adopt business software or build a tool? A decision method and an order-tracking example to prepare your rules, data and migration.
Read the article
First week of beta: what broke, what we fixed
Our first tester never saw the second screen. An honest account of the first days: three real defects, what they taught us, and what remains open.
Read the article
How much does application development cost in 2026: agency, freelancer or AI?
An agency charges a median €30,000, a freelancer €8,000 to €20,000, and an AI agent team a few dozen euros per month. The full picture: what each option buys, and when each remains the right choice.
Read the article
Quality, measured: six days of duels against the reference method
We promised that one day we would measure more than restraint. We have: over 180 jury verdicts, twelve method versions, an initial defeat, and a lesson about measurement itself.
Read the article
Your specifications will soon be worth more than your code
AI makes code abundant. Understanding becomes the scarce resource: knowing what you wanted to build, and why. Three levels of maturity are emerging.
Read the article
Agency, freelancer or AI tool: how to choose when launching your product
You have a product idea and no technical team. Three paths are available, each with its own logic; the right one depends on how much control you want to keep.
Read the article
Application requirements: a guide, example and free template
Describe your needs without technical vocabulary: users, journeys, rules, data and acceptance criteria. A booking example and a free text template to adapt to your project.
Read the article
Directing AI agents without knowing how to code: what has changed
AI has been able to write code for a while. What is new: a non-technical person can now direct it.
Read the article
What our ×10 measures, and what it does not
We publish one figure: around 10 times less methodological context injected into agents. Here is the full measurement, including its limits.
Read the article
The deliberate opposite of a terminal
Why Maestro looks like a concert hall rather than a black screen: the setting is part of the product.
Read the article
Why nine agents have first names
Margaux, Victor, Salomé, Maurice, Constance, Marcel, Amélie, Félix and Léa. Names that make the work easy to follow.
Read the article