Journal: behind the scenes

A whiteboard covered with coloured labels in an evening office, a brass lamp lit beside a mug

October 13, 2026 · 7 min read

Behind an application dashboard: showing that a product is running

Our second tester manually moved four steps on their board because the application no longer did it. On August 26, we rebuilt how Maestro shows its work: ten workstreams, 43 defects fixed, and a lesson for every dashboard.

An empty chair in front of a conductor's stand in a rehearsal room at the end of the day

September 27, 2026 · 5 min read

Behind an application update: a signed-out assistant no longer goes unnoticed

A tester tells us they installed their AI assistant without ever signing in. The software treated 'please sign in' as completed work. Here are the three fixes released in version 0.1.17, and the two defects we are leaving open until the next pass.

A repair bench in the evening: a hand adjusts a watchmaker's loupe over annotated pages beneath a brass lamp, with a laptop and its blurred screen further back

23 September 2026 · 11 min read

Maestro beta: request access and prepare your first project

How to request an invitation, prepare your AI access and choose a useful first trial. A guide to exploring Maestro on Mac with a clear goal and realistic expectations.

Two chairs facing each other in a home office at the end of the day, an open notebook covered in blurred notes on an armrest

23 September 2026 · 12 min read

Validate your app idea with Margaux before building

Between an appealing idea and a confirmed problem, there is often some research to do. Prepare your observations, work on the brief with Margaux and decide what to check before building.

behind the scenesvibe coding

A dismantled engine on a garage workbench, parts lined on a cloth and a lit work lamp

September 23, 2026 · 6 min read

A free AI assistant delivering code that did not compile: three tuning passes

We wanted a route without a subscription. We put a free assistant on our test bench, and it delivered twenty-five approved stages whose code would not launch. What three tuning passes taught us, about it and especially ourselves.

A meeting room emptied afterwards, with a bound document open between two cold cups and a pushed-back chair

September 21, 2026 · 5 min read

Why software projects fail despite requirements: silent disagreement

The document was complete, reviewed and approved by everyone. Nine months later the product launched and nobody wanted it. What these stories reveal about obtained versus real agreement, and what an approval gate must provoke.

behind the scenessecurity

A set of keys forgotten on a locksmith's counter, with a drawer ajar behind it

September 18, 2026 · 6 min read

API key exposed in an application: an autopsy of a defect, and $600,000 at METR

A tester sent us their project because its online database refused to connect. Opening it, we saw their access key in plain text inside the application. What that costs elsewhere, and what we changed the same day.

A brass reception bell beside an open notebook at a hotel reception desk in the evening

September 13, 2026 · 6 min read

Behind the scenes: the bell that does not shout, rebuilding an application’s notifications

Our first useful notification arrived drowned in everything requiring nothing. Five comments in one session, one simple cause, and a rebuild delivered the same day: the story of an avalanche we created ourselves.

The conductor's stand, with a baton resting on the score before the musicians' empty chairs

September 3, 2026 · 5 min read

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.

pricingbehind the scenesvibe coding

A small, modest desk in the foreground beneath a brass lamp, with a much larger, more luxurious office glimpsed blurred through an open door

September 1, 2026 · 6 min read

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.

A workshop pegboard covered with dozens of labelled self-service tools, a hand taking a screwdriver and a small unattended cash tray underneath

August 30, 2026 · 6 min read

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.

behind the scenessecurity

A night inspection: a hand holding a red pen rereads pages covered in circles and corrections, with a blurred screen showing lines of results in the background

August 29, 2026 · 5 min read

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.

A hand sorts receipts one by one beneath a brass lamp, beside an open account book covered in handwritten figures

August 28, 2026 · 3 min read

What do our AI agent trials cost? A campaign summary

One internal campaign covered eight trials and the first set of features for each project. Here is what its $105.64 includes, and why that is not the price of an application.

pricingbehind the scenes

A hand gently places a handwritten card on a stack of approved cards beneath a brass lamp, a calm gesture of acceptance

August 26, 2026 · 6 min read

Before approving an AI proposal, what should you check?

A document can sound convincing and miss an important rule. Three things to check so your approval really matches what you need.

An old cast-iron and brass scale on a desk, with a hand adding a weight to one pan to recalibrate it and no screen in the scene

August 25, 2026 · 3 min read

When the score falls: what our software comparisons showed

In August 2026, a newer version received a lower average score. We compared the software outputs in pairs to understand the result. Here is what that check showed, without converting the verdicts into a new score.

businesssecuritybehind the scenes

A hand locks the wooden drawer of a therapist's desk, with two armchairs facing each other in the background and no computer in the room

August 24, 2026 · 3 min read

Therapists: choose a tool without exposing session notes

Finding information between appointments should be simple. Before building a tracking tool, decide what it holds, who can access it and how that information will be protected.

behind the scenescomparisonsmethod

Two small handwritten cards on a desk, one covered in ticks and the other in crosses, with a miniature brass trophy placed exactly between them, on neither card

August 23, 2026 · 3 min read

Why we also describe failures in our AI agent tests

A good score can hide a missing feature. Two results from our internal trials show why we need to examine errors and specify what we actually measured.

methodcomparisonsbehind the scenes

A woman directs an AI agent team from her desk in the evening, with Maestro's tracking board on screen

August 23, 2026 · 3 min read

Can a team of AI agents deliver software? Our tests and their limits

Our tests covered an initial development batch. Here is what they showed, how we evaluated it and what still needed checking before real use.

comparisonsmethodbehind the scenes

A rehearsal room emptied at night: brass music stands folded against the wall, an open moving box, and one new stand left alone beneath a pendant light

August 22, 2026 · 3 min read

Why Maestro moved away from BMAD: our decision and test results

We replaced BMAD with our own method to organise Maestro’s steps more directly. A campaign of nine scenarios in August 2026 informed that decision without establishing a general ranking of tools.

A test chair beneath a lamp in the evening, a stopwatch on the seat, a broken cup and bent paper clip on the floor: traces of testing pushed to breaking point

August 19, 2026 · 3 min read

How an AI agent helps us test Maestro’s responses

A user may reply too quickly, request a correction or reverse a decision. Our test setup simulates these situations to look for blockages, within clearly defined limits.

August 18, 2026 · 4 min read

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.

methodcomparisonsbehind the scenes

Two stacks of printed documents compared line by line in pencil beneath a brass lamp

August 10, 2026 · 3 min read

Six days of testing: what we measured against BMAD

From August 5 to 10, 2026, we compared versions of our method with BMAD and then with each other. The results concern documents and an initial set of features, not complete applications delivered to customers.

behind the scenescomparisons

An old kitchen scale beside a laptop showing Maestro's kanban board, with a hand taking notes

July 27, 2026 · 3 min read

What our ×10 measures: instruction text, not your bill

Our historical measurement compares the instruction volume of two methods. It does not mean ten times cheaper or ten times better. Here is the calculation and its scope.

A man closes a laptop casting a green terminal glow in a warmly lit room

July 27, 2026 · 4 min read

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.

An empty rehearsal room, with nine chairs and music stands in a semicircle in the light from a window

July 27, 2026 · 3 min read

Why nine agents have first names

Margaux, Victor, Salomé, Maurice, Constance, Marcel, Amélie, Félix and Léa. Names that make the work easy to follow.