Navigate
HomeStart here
MusingsResearch & long-form
BuildingProjects & learnings
WorkProfessional practice
RunningTraining & races
AboutValues & identity
Life & PlacesCulture, food, travel, cities
Notes & ArchiveJournals, essays, portfolio
Workflow essayTechnology & IntelligenceUpdated April 2026Architecture synthesis + product framing
April 2026Workflow essayTechnology and intelligence

Jenn OS should be the operating layer, not another chat skin.

The strongest lesson from the recent wave of multi-model tooling is structural. One durable method, many wrappers, local records, repairable setup, and a clear path from today's work to tomorrow's context. That is the shape Jenn OS should take.

v1 should not try to be a universal personal operating system. It should ship one narrow but valuable loop: open the day with local truth, prepare the right context before work starts, keep the active contract visible while building, and close the session into durable memory.

V1 scope
Morning brief from open loops, logs, and active repos.
Project context packs before the agent touches work.
Build support that keeps the output contract in frame.
Closeout that writes memory at the correct layer.
v1 stages
4
Open, route, build, close.
core wrappers
4
CLI, skill, MCP, review surface.
memory rule
1
The worker writes the record.
hidden telemetry
0
The source of truth stays local and explicit.
01

Why another layer at all?

Context drifts between sessions.

The important facts already exist, but they live across build logs, workspace notes, active branches, transcripts, and site docs. Chat alone does not keep that system coherent.

Plans do not naturally close.

The hard part is not generating ideas. The hard part is ending a session with a trustworthy record of what changed, what is canonical, and what should happen next.

Agent capability decays without repair paths.

A skill that cannot be installed, checked, and repaired locally will eventually collapse into one-off setup folklore. Productizing the wrapper matters as much as productizing the idea.

02

The daily loop should be small and real.

1 / Open

Morning brief

Start from local truth rather than from whatever happens to be in the current chat window.

Open loops from build logs and pinned commitments
Active repo and project state
What changed since the last touch
2 / Route

Context pack

Prepare the right project before work begins, not halfway through the task.

Relevant `_WORKSPACE.md` and `_BUILD_LOG.md`
Skill files and canonical commands
Expected output and closeout path
3 / Build

Working surface

Keep the active contract visible while the work is moving so the agent does not drift.

Objective and success condition
Loaded sources and quality checks
Decision lane when a real tradeoff appears
4 / Close

Scoped memory

Treat closeout as part of the product rather than as an optional afterthought.

Build log draft from real artifacts
Canonical output recorded
Tomorrow hook and skill promotion cue
03

What Farness clarified

Max Ghenis's Farness project is nominally about decision-making, but the reusable lesson is more general. The durable thing is not the plugin. It is the product shape underneath the plugin: one stable method, a local store, a package-first setup path, an MCP surface, and commands that can inspect and repair local drift.

That shape translates cleanly to Jenn OS. Keep the method stable. Store durable objects locally. Expose the same system through CLI, skills, and MCP. Add a doctor path early. Let the UI compile from explicit records instead of pretending the interface is the memory.

Figure 3A

System shape

The product shape is simple on purpose. Many wrappers at the edge. One operating loop in the middle. Explicit local records underneath. If the page disappears, the system still exists.

CLI
Skills
MCP
Pages
Step 1
Open
Step 2
Prep
Step 3
Build
Step 4
Close
_BUILD_LOG.md
_WORKSPACE.md
Project docs
Design rule

The wrappers should feel native, but they should not own the method.

The loop should stay stable even if the model, shell, or plugin changes.

The records should remain readable even if the UI goes away.

Figure 3B

Farness translated

Farness pattern
Stable method
Jenn OS translation
Open, prep, build, close

The durable thing is the loop, not the host surface.

Farness pattern
Local store
Jenn OS translation
Scoped logs and workspace docs

Memory stays inspectable and close to the work.

Farness pattern
Package-first setup
Jenn OS translation
`setup`, `doctor`, `uninstall`

Repair is part of the product, not an afterthought.

Farness pattern
Many entry points
Jenn OS translation
CLI, skills, MCP, pages

One system should feel native in more than one shell.

Method
What stays constant
Jenn OS should anchor on one operating rhythm: open work, prepare context, build against a visible contract, close out into durable memory. The wrappers can change later.
Objects
What survives chat
Open loops, session summaries, context packs, decisions, and scoped build-log entries should all be plain local records. Inspectable beats magical.
Wrappers
How it feels native
A small CLI, packaged skills for Codex and Claude, and an MCP layer should all point at the same local state. The package-first path matters because repair is part of the product.
View layer
What the page is for
The page should make today legible. It should not become the only place the system makes sense. If the UI vanished, the records and commands should still be readable.
04

What v1 includes, and what it refuses

Include

Morning brief

A ranked summary of open loops, active projects, and review debt from local truth.

Prep command

A context pack builder that loads the right files and rules before the task starts.

Closeout draft

A structured end-of-session summary grounded in changed files and scoped memory rules.

Repair path

Commands like `setup`, `doctor`, and `uninstall` so the system can survive tool drift.

Refuse
No hidden vector-memory religion pretending to replace explicit local records.
No universal life dashboard before the work loop itself is trustworthy.
No plugin-first framing that makes the product depend on one host surface.
No autonomous theater where the interface looks alive but cannot explain itself.
05

The real point

The most interesting question is no longer whether multiple models can collaborate. They can. The more important question is where judgment, memory, and verification should live when the models, shells, and plugins keep changing. Jenn OS is my answer to that question.

Bottom line

Jenn OS should make the daily loop legible, durable, and repairable. If it cannot do that, it is not an operating layer yet.

That is why v1 matters. A narrow local operating layer can compound. A vague personal-OS fantasy usually turns into a prettier version of the same forgetting.