# mornati-developer-coding-agents-bmad-experiment-2026-03-14

## Veille

Developer's role in the face of AI coding agents, one-day BMAD method experiment, evolution toward agent supervisor - Technical blog

## Titre Article

What is a Developer When We Use Coding Agents? My 1-Day BMAD Experiment

## Date

2026-03-14

## URL

https://blog.mornati.net/what-is-a-developer-when-we-use-coding-agents-my-1-day-bmad-experiment

## Keywords

coding agents, BMAD, developer, agent supervisor, 70% problem, Agile AI, AI personas, software architecture, developer productivity, code quality, role evolution, junior developer, product manager, orchestration

## Authors

Marco Mornati

## Ton

Profile: practitioner-experimenter perspective, narrative-analytical register, intermediate-advanced level. Marco Mornati adopts a personal and reflective tone, recounting his one-day experience with the BMAD method as a structured narrative. The style blends concrete hands-on feedback (screenshots, tool choices, results obtained) with strategic reflection on the future of the developer profession. The author is honest about the limitations encountered: architecture remains a point where human expertise is indispensable. The tone is pragmatic, neither euphoric nor alarmist, with figures backing up the observations. Target audience: developers, tech leads, software architects, technical managers.

## Pense-betes

- BMAD = Breakthrough Method for Agile AI-Driven Development: applies Agile discipline to AI-assisted coding with specialized personas
- 5 BMAD phases: market research, analyst (PRD), epics/stories, technical architect, UX
- The project idea strengthened over the course of the AI-guided process (PRD)
- Use of Stitch (Google AI) for UX to avoid the generic AI aesthetic
- Use of gemini-cli for the discovery phase to save tokens for development
- Architecture is the point where human technical skills remain absolutely necessary
- Market statistics: 93% of developers use AI assistants, 27% of code in production is AI-generated
- Coding is 26% faster but delivery only 8-10% faster: the bottleneck is not writing code
- The "70% problem": AI reaches 70% quickly, the remaining 30% requires deep expertise
- Unsupervised AI code introduces 1.7x more defects
- Role evolution: Developer -> Software engineer -> "Agent supervisor"
- Gartner predicts that by 2028 the developer's role shifts from implementation to orchestration
- The junior developer's dilemma: how to learn without hands-on coding practice?
- The product manager role is under threat: technical profiles can take on the product preparation phase

## RésuméDe400mots

Marco Mornati recounts his one-day experiment with the BMAD method (Breakthrough Method for Agile AI-Driven Development), a methodological framework that applies Agile rigor to AI-assisted development. The method relies on specialized AI personas (analyst, product manager, architect) and unfolds in five phases: market research, drafting a PRD (Product Requirements Document), breakdown into epics and stories, technical architecture, and UX design.

Mornati describes how his project's initial idea progressively strengthened and clarified over the course of the AI-guided process, notably thanks to the PRD phase. For the UX part, he used Stitch (a Google AI tool) to avoid the generic aesthetic typical of AI-generated interfaces. For the exploratory phase, he opted for gemini-cli, reserving his more costly tokens for the actual development phase.

The central lesson of the experiment is that software architecture remains the domain where human technical expertise is absolutely indispensable. AI excels at generating code, but architectural decision-making requires a deep understanding of trade-offs, system constraints, and long-term implications.

Mornati supports his reflection with market data revealing a paradox: while 93% of developers use AI assistants and 27% of code in production is now AI-generated, coding is only 26% faster and actual delivery only 8-10% faster. This gap illustrates that the bottleneck is not writing code but understanding, architecture, testing, and integration.

The article highlights the "70% problem": AI enables reaching 70% of the result quickly, but the remaining 30% - complex debugging, edge cases, system integration - demand deep human expertise. Moreover, unsupervised AI code introduces 1.7 times more defects, underscoring the critical importance of human review.

Mornati traces an evolution of the profession: from developer (one who writes code) to software engineer (one who designs systems) toward the "agent supervisor" (one who orchestrates AI agents). Gartner predicts that by 2028, the role will shift from implementation to orchestration. This transformation raises two major questions: how will junior developers learn their craft without hands-on coding practice, and what future awaits product managers when technical profiles can now take on the product preparation phase thanks to AI tools?

## GrapheDeConnaissance

- Marco Mornati —utilise→ BMAD (METHODOLOGIE, 0.99)
- BMAD —est_basé_sur→ discipline Agile (CONCEPT, 0.95)
- BMAD —utilise→ personas IA specialisees (CONCEPT, 0.95)
- Marco Mornati —utilise→ Stitch (TECHNOLOGIE, 0.98)
- Marco Mornati —utilise→ Gemini CLI (TECHNOLOGIE, 0.98)
- Architecture logicielle —utilise→ expertise humaine (CONCEPT, 0.95)
- Code IA non supervise —mesure→ 1,7x plus de defauts (MESURE, 0.9)
- Gartner —prédit→ le role du developpeur passera de l'implementation a l'orchestration d'ici 2028 (AFFIRMATION, 0.88)
- Superviseur d'agents —remplace→ Role developpeur traditionnel (CONCEPT, 0.9)
- IA —mesure→ 70% du resultat atteint rapidement, 30% restants exigeant expertise (MESURE, 0.92)
- Marco Mornati —affirme_que→ l'architecture est le point ou l'expertise humaine reste indispensable (AFFIRMATION, 0.93)
- Profils techniques avec outils IA —concurrence→ Role chef de produit (CONCEPT, 0.85)
- Codage IA —mesure→ vitesse de codage +26% (MESURE, 0.9)
- Livraison logicielle —mesure→ acceleration de seulement 8-10% malgre l'IA (MESURE, 0.88)

---
Canonical: https://www.thekb.eu/en/fiches/mornati-developer-coding-agents-bmad-experiment-2026-03-14/