Sérgio Tavares
The scene repeats itself in every company right now. A designer fine-tunes a Claude prompt for brand tone. A recruiter trains a CustomGPT for job descriptions. A data analyst builds a workflow to summarize meetings. Each is doing something powerful—creating reusable digital artifacts—but almost none are sharing them in ways others can find, trust, or build on. What should be a commons becomes a closet. The result: dozens of “hero projects” running in parallel, each magical in isolation, none sustaining momentum.
At home, we see a similar pattern. Parents test new apps to track kids’ screen time or filter content. Each family crafts its own governance system from scratch. Small victories, local logic, no shared playbook. We live in a time of personalized wizardry—and very little coordination.
Why it matters: when every expert becomes a mini product team, the organization risks losing coherence, accountability, and trust in shared tools.
The governing idea
Building with AI no longer means using tools; it means creating them. That shift moves every expert from contributor to product owner—responsible not just for the work, but for how that work scales, teaches, and integrates. The missing skill isn’t more prompting or fine-tuning. It’s service design: making one’s creations legible, maintainable, and adoptable.
Three truths about the new builder’s landscape
- Artifacts without pathways die.
- Craft owners must speak in service logic.
- Governance is creative, not bureaucratic.
- Purpose: What problem does this solve?
- Dependencies: Which data, APIs, or teams are touched?
- Reusability: What part can others safely reuse or fork?
A style sheet for an LLM or prompt library only matters if others can find, trust, and remix it. The company needs a shared registry for these assets—part library, part directory. Think “internal npm for prompts.”
Builders must define who benefits, under what conditions, and how feedback loops will work. This is product ownership in miniature. Without that framing, every experiment risks becoming folklore—known by a few, lost at reorg.
The goal isn’t to slow innovation but to create visibility. A lightweight “AI Project Canvas” can require three things before launch:
When every project answers those, it becomes easier for operations, legal, and leadership to track both opportunity and risk.
Design the path
Principles: Transparency, reuse, collective learning.
Patterns: Internal registries, AI canvases, short “artifact demos.”
Journeys: From individual tool → shared template → approved reusable component.
Metrics: Number of reused artifacts; ratio of duplicated to shared projects; peer adoption rate.
Governance: Monthly cross-domain “show & share” reviews where builders present what they made and others nominate for inclusion in the shared catalog.
Quick pilots:
- Run a one-week “Artifact Review” where teams submit their AI experiments into a central doc. Note overlaps and missing links.
- Launch an “AI Builder Guild” (1 hour/month). Keep it social—less gatekeeping, more show-and-tell.
- Create a “Prompt Pattern Library” in Notion or GitHub; include versioning and metadata (owner, purpose, dependencies).
Obstacles & how to unblock
- Legal anxiety: Teams fear exposing sensitive data or IP.
- Incentive misfit: Builders are rewarded for novelty, not reuse.
- Cultural heroism: Experts identify with autonomy.
→ Start with public-domain examples; add “clean prompt” templates.
→ Add recognition for shared artifacts in performance reviews.
→ Frame reuse as craft elevation, not control. Make the commons aspirational: “Your tool made our culture smarter.”
Evidence & citations
Research on digital workplace overload shows that fragmented tech efforts lead to cognitive and operational burnout (Marsh et al. 2024). Collaboration improves when artifacts are legible and shared (Latour 2005; Hui 2019). Service-design methods—when applied internally—reduce redundancy and improve trust (IDEO 2023).