I’ve gradually come to see that the projects I’m working on, which look scattered from the outside, are actually responses to the same underlying question: as AI begins to change the conditions beneath work, education, business, and knowledge circulation, some of the interfaces that used to connect people are becoming inadequate.

This is a layer deeper than asking which jobs AI will replace. The more fundamental change, as I see it, is that many of the mechanisms we’ve relied on are starting to show cracks — schools feeding graduates into the workforce, entry-level jobs training new people, CVs and job descriptions helping talent and companies understand each other, small businesses finding technical services through traditional channels, managers learning about technology transformation through consultants and briefings, communities accumulating knowledge through articles and word of mouth. None of these mechanisms were perfect before. But they largely worked. After AI, they won’t disappear overnight — they’ll just start showing more fractures.

Agent Mindset: language and framework

My book on AI agents is the starting point for this thinking. On the surface, it’s an attempt to break down what an AI agent actually is: models, context, tools, memory, multi-agent systems, governance, organization. But what I’m really trying to work through is what happens when AI shifts from being a tool to being a system you can delegate tasks to — and what that means for how people and organizations need to think about work, judgment, and responsibility. The book offers a vocabulary for not just being impressed by AI’s capabilities, but being able to see the structure behind them.

Orbita: an agent-native infrastructure experiment

If the book provides language and framework, Orbita is where I’m testing those ideas as working infrastructure. Most agent products lead with an interface designed for humans: a chat window, an IM channel, a web UI. Orbita inverts that. Its primary interface is an API — designed for how other agents or orchestrators call an agent runtime, rather than how a person converses with one.

That’s a technical choice, but it follows from a judgment about how websites and knowledge systems are likely to work: more and more reading, writing, organizing, submitting, and tracking won’t be done manually by people in forms, but by people’s agents, through APIs, on their behalf. Orbita is an early experiment in that direction.

From infrastructure to specific social interfaces

From here, I started thinking about a few more concrete problems.

The first is talent formation. AI may be quietly eroding the training function that entry-level work used to serve. Junior employees were cheap enough relative to their output that companies could afford to train them alongside the real work. AI has shifted that calculation. More importantly, entry-level work provided something harder to replace than income: an environment where you learned to scope ambiguous problems, verify your own judgment, know when to ask for help, and defend your choices to someone whose opinion actually mattered. The apprenticeship project I’m running is trying to rebuild that part — giving people the chance to practice, in real projects under real supervision and with manageable stakes, the judgment moves that still matter in an AI-assisted world.

The second is capability signaling. As work changes faster, compressed documents like CVs and job descriptions are getting worse at expressing what someone can actually do — or what a role actually needs. Powerhouse tries to rebuild that understanding from both sides: candidates express their capabilities through in-depth interviews, concrete examples, and continuously updated profiles; companies describe roles, requirements, and context with the same depth. The goal isn’t just better matching — it’s rebuilding a signaling system where capability and need can be seen, verified, and trusted.

The third is the gap between need and delivery. Many small business owners know exactly where their processes hurt. They don’t know whether what they need is a RAG system, workflow automation, an AI assistant, or just some clarity about their existing processes. AI Business Life is aimed at the stage before the need has taken shape — using conversation to turn a vague pain point into a spec draft that can be discussed, validated, and delivered. It may also become another path to real-world practice for apprentices and junior builders.

The fourth is transformation knowledge and community. AITransformation.io and AITransformation.org are the knowledge infrastructure for this part.

.io is aimed at the individuals inside organizations who are carrying responsibility for AI transformation: CEOs, CTOs, department heads, HR leads, managers, internal champions. The goal is to move beyond a generic content site toward a personal cockpit — where users get briefings, suggestions, and agent-accessible context calibrated to their role, industry, project stage, and organizational state.

.org is aimed at a broader community of practitioners. Not a forum, and not just a publishing platform — more of a knowledge commons for AI transformation, where people can share field notes, failure experiences, events, questions, mentorship requests, and project requests. But longer-form contributions don’t have to be written manually inside a web form. The more natural model is for people to work through ideas with their own agents first, let the agent help structure and draft the contribution, and then submit or update it through an API. The site handles presenting content, tracking contributions, surfacing community highlights, and managing lightweight interaction — while keeping people in the position of judge and approver.

What these things share

If I had to summarize in a sentence: these projects are trying to build AI transition infrastructure. Not a single product, not a single business model, but a set of connected experiments. The book provides a worldview and vocabulary. Orbita tests agent-native infrastructure. The apprenticeship project cultivates the next generation of capability. Powerhouse rebuilds capability signals. AI Business Life connects SME needs with AI builders. AITransformation.org and .io keep knowledge, experience, benchmarks, community interaction, and agent-mediated participation in circulation.

I don’t think any of these are mature yet — most are still early experiments. But they share a belief: that the disruption AI brings won’t be smoothly absorbed by markets and institutions on their own. Some old interfaces will thin out. Some paths will break. Some capabilities will become more important while becoming harder to see. If we don’t want to learn everything only through hard lessons, we’ll need to deliberately rebuild some new interfaces — so that people can be developed, capability can be seen, need can be translated, knowledge can circulate, and people inside organizations can make better judgments.

Where this line goes, I’m still working out.

---

Project index

• Agent Mindset — A book about AI agents, system design, organization, and where human judgment belongs.
• Orbita — get-orbita.com — An agent-native, API-first agent system exploring how other agents and orchestrators call an agent runtime.
• AITransformation.io — ai-transformation.io — A personal AI transformation cockpit for individuals inside organizations, with a public library and a logged-in layer for personalized briefings, assessments, and knowledge context.
• AITransformation.org — ai-transformation.org — A knowledge commons for AI transformation practitioners, including field stories, learning guides, community interaction, apprenticeship, and agent-mediated participation.
• AI-Era Apprenticeship — jackyma.info/blog/ai-era-apprenticeship — Rebuilding the mentorship and real-project training that entry-level work used to provide.
• Powerhouse — powerhouse.zeabur.app — A high-resolution talent and role matching system using deep profiles, interview agents, and agent-to-agent verification to rebuild capability signals.
• AI Business Life — ai-business.live/en — Connecting SME owners with AI builders by turning vague pain points into discussable, verifiable, deliverable spec drafts.