Dedicated to the HR managers who patiently spent years explaining to me that I needed to focus and pick one thing.
There is a term — "scanner."
Barbara Sher popularized it in Refuse to Choose! A scanner is a person who finds it hard to spend a lifetime on a single thing. Not because lazy. Not because irresponsible. Not because incapable of finishing.
Because the mind is wired differently.
A scanner is interested in many things. Catches fire on ideas quickly, sees connections between fields, constantly finds new directions, new opportunities, new branches. The brain refuses to walk a single narrow road. It scans the horizon all the time.
Related concepts. Multipotentialite — a person with many abilities and interests. The more dismissive label is shiny object syndrome. There are also clinical topics like ADHD, but that is a separate story; mixing them all together helps no one.
I am talking specifically about scanners.
About people who feel cramped inside one task, one project, one profession, one role.
And for a long time this is treated as a problem.
You hear:
"Focus."
"Pick one thing."
"Stop scattering."
"Finish this first."
"You are jumping again."
"Nothing will come of it."
In some sense they are right.
If a scanner truly hops between ideas with no fixation, no architecture, no system — he can spend years in a permanent kickoff state.
But I am increasingly convinced: the problem is not that the scanner switches.
The problem is that he doesn't have a system big enough to make switching useful.
A small goal kills a scanner
When you give a scanner one small task and say "stay on this," he runs out of energy fast.
Not because the task is bad.
His mind doesn't have enough room.
He starts suffocating. Wants to switch. Feels caged. Glances toward other ideas. From the outside this looks like lack of discipline.
But maybe a scanner doesn't need a small task.
He needs a big goal.
Big enough to contain many projects, roles, directions, levels of difficulty.
Then something interesting happens.
The scanner still switches. But he no longer falls out of focus. He doesn't run away. He moves inside one large system.
Today he works on a landing page.
Tomorrow — on the architecture of an identity provider.
Then on DevOps and self-hosted runners.
Then on an agent that reads incoming mail.
Then on a search engine with hybrid ranking.
Then on content, brand, documentation, monitoring, business model.
From the outside it still looks like "jumping."
But inside the big system it is no longer chaos.
It is patrolling the territory.
I took on too big a project
I did exactly that.
Bought the maximum Claude Code subscription, leased five servers for roughly the same money, and started building not a single product but a whole ecosystem — Arcanada.
From the outside this looks insane.
"It's a fantasy."
"You won't pull it off."
"Pick one thing."
"You are scattering."
"You are just playing with AI."
I understand why people think that.
One person, limited budget, many directions, around thirty domains, several apps, agents, infrastructure, knowledge bases, trackers, automation.
Too much of everything.
But in three-plus weeks I have already pushed most projects to a working state. Some are live in production. Some accept registrations. Some are still in design. And in several places agent autonomy has reached L2–L3 on our internal AAL scale — an autonomy scale from L0 (manual) to L5 (fully autonomous with hard cost ceilings).
The 30-day subscription has not yet expired.
And the strangest part: most of these weeks were not spent on the products themselves. Those I built quickly. The time was eaten by the thing without which none of them would have flown — my own work-management framework that every task and every agent passes through. I call it Datarim.
The ecosystem map
To make the scale concrete, here is what currently lives under the Arcanada umbrella.
Identity and infrastructure.
- Auth Arcana (
auth.arcanada.ai) — the single identity provider for the whole ecosystem. OIDC, OAuth 2.1, Passkeys, federation with ten-plus providers, ReBAC via OpenFGA. No new service ever creates its own user table. - HashiCorp Vault — secrets, auto-unseal via Vault Transit, Tailscale-only access. After any reboot the service comes up with no operator.
- Tailscale mesh — private network across seven devices; all internal ports ride it.
User-facing products.
- Verdicus (
verdicus.app) — macOS assistant: translation, styling, visual analysis. Swift + NestJS API. - Munera Arcana (
muneral.com) — task tracker for AI agents. API already in production. - Arganize.me — Telegram life assistant.
- Transcribator (
transcribator.com) — commercial transcription platform, Groq Whisper under the hood. - Combateka, Initiative 10, Foodamunition Group and a handful of other landings and sites.
Internal agents and services.
- Datarim — the workflow framework I run every task through. Roughly thirty skills, dedicated commands per stage: PRD, plan, implementation, QA, review, publishing.
- Scrutator — retrieval and meaning engine: BGE-M3, ColBERT, hybrid search. Already indexes more than a thousand fragments from the internal knowledge base.
- Long Term Memory — persistent memory for agents. Hindsight as the benchmark winner.
- Model Connector (
connector.arcanada.ai) — unified API on top of Claude, OpenRouter, and CLI agents. - Agent Dreamer — the night-time librarian for the knowledge base: organizes, deduplicates, weaves cross-references.
- Ops Bot (
ops.arcanada.ai) — bidirectional Telegram agent for ecosystem operations. - Support Center (
support.arcanada.ai) — support with OAuth. - Disk Arcana — open-source file syncer for knowledge bases, written in Rust.
- Email Agent, Inbox Organizer, Screen Reader, Twitter Collector — small agents, each closes a slice of routine.
Around thirty domains. Five servers: WWW (static and landings), PROD (NestJS backends in Docker), DB (Postgres + Redis + Mongo + retrieval APIs), AI (agents and email cron) and DEV for autonomous Claude work. Twenty repositories in the Arcanada-one org. Ecosystem-wide autonomy averages L1.8 right now; the target is L4 for production agents.
And one person plus a distributed crew of AI agents drives all of this.
The longest part is not the products — it is the lathe
If anyone asked where most of my time went this month, I would not name any of the products listed above.
Most of my time went into Datarim.
This is not marketing dressing, it is a working tool. A pipeline from idea to archive: brainstorming, PRD, design, plan, implementation, QA, review, publishing, reflection. Every stage has its own command and its own agent. Eighteen commands, seventeen agents, around thirty skills.
This whole superstructure exists for one purpose. So that one person can move ten projects in parallel and not drown in his own chaos.
The thing is, you always want to make products first. Things that show. Sites, agents, features. But without the lathe those products come out half-assembled and rot fast.
So most of these weeks I was building the lathe. Fixing it, reshaping it, adding rules, teaching agents to review each other, rewriting plan templates, debugging auto-checks, running regressions. Boring, invisible, foundational work.
Cold number. Between April 10 and today, 266 tasks have closed through Datarim. Product features, infrastructure, framework upgrades. Each task with its own archive, its own reflection, its own decision history.
This is the "big system inside the big system" — the frame that turns scanner-style switching into something productive instead of destructive.
AI agents change the economics of a scanner
A scanner used to hit a physical limit.
You can imagine ten projects, but you cannot simultaneously be a developer, an architect, a designer, a DevOps engineer, a product manager, an analyst, an editor, a tester, a manager, and a researcher.
You can switch, but every switch is expensive.
Context drains.
Documentation never gets written.
Code is forgotten.
Decisions scatter.
Projects start to rot.
This is exactly what AI agents rewrite.
They don't cancel the work. They don't do magic. They don't turn chaos into a product automatically.
But they let a scanner build a working system around himself, where some of the context lives in external memory, some tasks delegate, some processes automate, some reviews go on a conveyor, and several projects move in parallel.
For me this turned into a concrete schema. Datarim drives a task through stages: brainstorm → PRD → plan → implementation → QA → review → publishing. Scrutator searches the knowledge base. Long Term Memory remembers how the previous iteration ended. Model Connector picks the model for the load. Ops Bot pings Telegram when something breaks. Auth Arcana hands every service scoped tokens. Vault doles out secrets.
Scanner-ness stops being a weakness. It becomes a way of governing a complex ecosystem.
Because a scanner sees the connections.
How a landing connects to an agent.
How an agent connects to a tracker.
How a tracker connects to a knowledge base.
How the knowledge base connects to memory.
How memory connects to code review.
How review connects to CI/CD.
How CI/CD connects to autonomy.
How autonomy connects to product.
How product connects to market.
How market connects to content.
How content connects to community.
For linear thinking this is overload.
For a scanner it is the natural map of the world.
The secret isn't to stop switching
I am increasingly sure that a scanner doesn't need to forcibly turn into someone else.
No need to break yourself across the knee.
No need to try to become a "normal" narrow specialist.
No need to feel ashamed of being interested in too many things.
You need something else.
You need to build a frame so big that all your switches work toward one goal.
Not ten random projects. One ecosystem.
Not chaotic ideas. Connected modules.
Not escape from boredom. Context-shift inside a shared architecture.
Not "I dropped one and started another." But "I moved to a different part of the same system because that is where the energy is right now."
That is the substantive difference.
A scanner is destructive when he hops between unrelated ideas.
A scanner is incredibly powerful when he builds a big system where every new idea becomes not a distraction but a new module.
The big goal as a container for chaos
A scanner doesn't need motivation.
He needs a container.
A big goal is a container for his chaos.
Complex enough not to bore him.
Wide enough to host many directions.
Deep enough that there is always more to dig.
Coherent enough that switching does not break focus.
Ambitious enough that the brain stops trying to escape sideways.
For me that container became Arcanada — an ecosystem of autonomous AI agents and the products around them. It even has a slogan we put on every site: "One human life matters." This is not marketing copy. It is the frame that explains why all of this complexity even exists.
Inside it there are landings, servers, knowledge bases, trackers, agents, interfaces, automation, DevOps, content, business model, research, and the product's internal philosophy.
And I move among them without losing meaning.
Tired of the frontend — I go into Auth Arcana architecture.
Tired of architecture — into DevOps and runners.
Tired of DevOps — into texts for the blog.
Tired of texts — into the dreaming agent.
Tired of agents — into the product mechanics of Munera.
But all of it works on one system.
You feel like you are switching all the time, but you don't lose focus.
You are not in a single cell. You are inside one city you keep building yourself.
Maybe this is the right working shape for a scanner
Maybe a scanner doesn't need to pick one project.
Maybe a scanner needs to pick one big mission. And then build dozens of connected projects inside it.
Maybe the problem isn't an excess of ideas. It is the absence of an architecture able to carry those ideas.
And here AI agents become a critical amplifier.
They let one person work not as a soloist but as a small distributed team. Imperfect. Buggy. Under constant supervision. With rollbacks, reviews, rework, and the daily fight against hallucinations.
But even in that form it changes the scale of what is possible.
In three weeks one person can lift what used to look impossible for a team in months.
Not perfect. Not final. Not flawless.
But enough to see the outline of a new reality.
I don't know if it will work
I am not writing this as a success story.
Too early.
Not all projects have proven their business model. Not all are launched. Not all generate revenue. Not everything is stable. Not everything is in the shape I want. The ecosystem AAL average is L1.8; the target is L4. There is a long way to go.
But I already see the main thing.
Scanner-ness is not a defect.
Inside the right architecture it becomes the engine.
When you have a big system, switching stops being procrastination. It becomes a way to keep energy. A way to see connections. A way to avoid burnout. A way to push several directions at once that feed each other.
And maybe people like this — scanners, multipotentialites, people with wide attention and a hunger for new connections — receive an unusual advantage in the age of AI agents.
The world no longer demands they shrink to a single function.
You can build systems now.
Big. Complex. Alive. Connected.
And inside that kind of system a scanner can finally be himself.
Not break himself for focus.
But build the kind of goal whose natural switching is his form of focus.