What BLOM is

What BLOM holds

BLOM, from Boundary Layer Labs, is where your organization keeps what it builds, what's intended for it, and the work that changes it. People and AI agents do that work side by side, and BLOM records both.

Your organization

Its people and their roles, its PDLC (the process work moves through), and its standards (the rules work meets). For example, Harbor Labs.

Products and their items

Each product your organization builds, and the items below it: its capabilities, their features, and their components. Each item has a number, such as 1-2, and can have an owner. For example, Member onboarding › Invitations.

Specs

Read from a GitHub repository you connect, where you write them in Markdown beside your code. BLOM keeps each revision with the commit it read it from, and never writes to your repository.

Tasks

Jobs create or change items. Reports, such as Bugs and Issues, record what's wrong, and reach work only through a Job that takes them on. Every Job is in exactly one initiative. For example, the Bug "Invite link opens twice", taken on by the Job invite-expiry-build.

The record

Every change records who made it, a person or an agent with the person it answers to, and when and why.

The operating model behind these records is the BLOM method. It explains why BLOM keeps them apart.

How people and agents work in it

People sign in with GitHub or Google. Each person holds a role at the organization, a team, or a product: a Viewer reads the work there, an Editor reads and changes it, and an Administrator also manages it and who has access.

Each open task has a next action: the one step it waits on, and who it waits on. Your queue lists the tasks waiting on you, priority first, then longest waiting.

Agents, such as Claude Code and Codex, connect through BLOM's MCP server with a key that a person makes. BLOM records the agent's work under the key's name, with that person beside it, such as Claude · Ada Lovelace.

An agent finds the next actions waiting on it and claims the step it's working on, so everyone can see who holds it. It completes the step with one of the step's outcomes, then hands back a closure: what changed, the checks, and what's left. A closure is a handoff, not completion.

An agent's output grants no authority by itself, and no agent tool approves anything.

A tour of the rail

The rail down the left side of BLOM holds its places, in four groups.

Views

Atlas View

Your organization's home: counts across the whole organization, with Get started at the top until you close it.

Product View

One product at a time: its items, and the state of its work.

Work

Items to Jobs

For one product and one of its initiatives, which Jobs target which items.

Your queue

The tasks waiting on you, and the notes that mention you.

Design Lab

Approved designs

The approved design for each of your screens.

Draft

Screen designs still being explored, one version after another.

Brand Guide

Your brand, skins, and design system.

Archive

Designs set aside, kept whole so a decision can be traced back to them.

Approving a design approves it for nothing else.

Organization

At the rail's foot, Organization opens a panel of your organization's places:

PDLC

The phases work moves through, and the gate on each move.

Standards

Your standards by category, and which revision of each is in force.

People and access

Who's in the organization and their roles, and, for Administrators, its invitations.

Connected tools

The GitHub repositories BLOM reads.

API keys

Your keys, which connect Claude Code, Codex, and scripts to BLOM.

Token use

The tokens your organization's work used, by day or month and by model, with what they cost: as agents report them, people enter them, or your AI providers' records show them.

Connect Agent

Connecting an agent, and reaching BLOM from a terminal, a script, or CI through the CLI and the public API.

Settings

Your organization's settings, such as whether Teams are on.

Below Organization, your name opens Your profile, and Sign out is beside it. Your organization's name, at the top of the rail, switches between your organizations. Help, in the top bar, sends feedback or a request to support.

What BLOM doesn't do

• It never runs your agents. BLOM doesn't start, dispatch, or deploy them, and it runs no AI models of its own. Your agents run where you run them, and connect to BLOM to find their work and record it.
• It never changes your repository. BLOM reads GitHub with read access only, and makes no change there.
• It never counts one fact as another. An approval isn't proof that something was built, and a merge isn't a deployment.

Where to start

1. Create your organization, with its first product and its PDLC.
2. Add a product for each other thing you build.
3. Invite people to work in it with you.
4. Connect a repository, so BLOM reads your specs and your pull requests.
5. Connect Codex or Claude Code, so your agents pick up work.

To begin the way the method suggests, read Start with one Product and one Job.

Related