Cultivated AI: you don't program it, you grow it
Ready-to-download kits (personal, Jarvis and company) and the idea that AI grows through use and correction, not code.
Cultivated AI is a way of organizing what surrounds a model so an AI agent can work better with you. The model arrives ready; cultivation happens through context, tools, rules, examples, memory, evaluation, and feedback. When the system improves, it is not because the model learned on its own: its files and instructions were adjusted.
The idea applies to personal assistants, such as a Jarvis, and to agents used by businesses. Instead of starting from scratch with every conversation, you define a role, share the context it needs, and review the results. By recording errors and correcting the rules, the agent may repeat fewer mistakes and take on tasks with more autonomy, as its history and evaluations show it is ready.
To get started, choose a clearly defined role, gather useful information in files, and set rules about what the agent may do or should leave to you. Use it, record where it made mistakes, and review the files regularly. Memory alone stores context; improving behavior requires evaluating results and applying corrections. Cultivation takes work, and results still need to be checked.
The link opens the original content; some require a subscription.
Go further on INEMA.CLUB
More news
- 02/10Raio-X de Margem helps clinics and salons track losses and goalsThe dashboard compares results before and after, and tracks losses, targets, and attributable bonuses in a pilot with 1,302 clinics and salons in Batel.
- 01/10OpenAI Dots: analysis, criticism, and comparison with other toolsAn analysis explains what OpenAI Dots is, compares it with Muse, Grokbot, and Hermes, and gathers reported criticism and issues.
- 29/09San Martín — new visual concept and homepage in three languagesA new desk-style visual concept with sticky notes and folders, plus a homepage available in Portuguese, English, and Spanish.
- 29/09Codex + Claude in a video: how each one plans and reviewsA video presents the six levels of Use Both and explains each tool’s role, while also showing open courses and kits.
- 28/09Trade com Agentes: a 7-day study with GPT-6 AstraA study presents routines, task handoffs, and strategy with GPT-6 Astra, and explains why the result fell short of the S&P; it is not a recommendation.