Atlas Tradecraft · No. 001 · MMXXVI

AI work
your auditors trust.
Your team owns.

Atlas Tradecraft teaches the Interpretable Context Methodology — ICM — a numbered, auditable pipeline that turns one brief into vetted AI output, with human review built in. Every project ships as a portable packet any AI can continue. Productivity goes up. Output quality goes up. And every step is reviewable, by your people or your regulators.

A certified specialist embeds one-on-one with your employee, rebuilds their week as an ICM pipeline, and hands back the playbook. They keep using it after we leave.

Claim a free 1-hour session Try the method at icmproject.com See the method →

Built for leadership at

Fortune 500 firms in finance, biotech, healthcare, manufacturing, and retail — wherever AI output has to be reviewable, portable, and defensible, not just produced.

01 — The Gap

The tools are deployed.
The method isn't.

Most enterprises now spend seven figures a year on AI licenses, vendor implementations, and a flurry of town-hall trainings. And yet — quarter after quarter — the productivity gains arrive in pockets, not across the org.

A few power users become 10× operators. Everyone else stays at 1.0× — and the work they do ship isn't reviewable, isn't repeatable, and walks out the door with whoever made it.

i.
0%

Productivity

of paid AI seats are used weekly

Licenses sit dormant while leadership waits on ROI dashboards. No method means no momentum.

ii.
3 in 4

Quality

employees say AI output is inconsistent and unverifiable

Generic training shows the buttons. It doesn't teach the structured pipeline that makes output trustworthy.

iii.
< 8%

Auditability

of enterprises can audit how their AI output was produced

Chat logs aren't an audit trail. When risk and legal ask, “show me how this was made,” teams have nothing portable to hand over.

02 — The ICM Method

One brief in.
A whole pipeline out.

ICM — the Interpretable Context Methodology — is an AI-native framework where the folder structure itself is the agent. You describe a project in a paragraph. ICM auto-enriches it into goals, audience, and voice, then drafts a numbered pipeline. Each stage runs in its own scoped context, writes its own reviewable artifacts, and hands off to the next.

The middle stage is a human review checkpoint. The last stage exports a portable packet — Markdown plus a manifest, zipped — that any AI agent can pick up and continue. No vendor lock-in. No black boxes. Output your auditors can actually read.

  1. 01

    Auto-enrich

    One paragraph becomes a structured brief — goals, audience, voice, success criteria — before a single prompt is written.

  2. 02

    Numbered pipeline

    FRAMER → RESEARCHER → PRODUCER → REVIEWER → SHIPPER. Each stage scoped, each stage reviewable, REVIEWER is the human gate.

  3. 03

    Portable packet

    Export as Markdown + ZIP. Any AI agent — today's or 2027's — can pick up the packet and keep working. Your IP, your storage, your auditors.

ICM is Atlas's curriculum. The tool that implements it lives at icmproject.com — self-serve for individuals; Atlas is the embedded enterprise tier that teaches it across your workforce.

03 — The Engagement

Three phases.
One audit-grade pipeline.

Every engagement runs on the ICM pipeline. The phases below are how the customer experiences it — the stages in monospace are the artifacts your auditors get.

  1. 01

    Frame the work

    ICM · 00 INTAKE · 01 FRAMER

    Your specialist shadows the employee, audits their tools and recurring artifacts, and writes the project brief — goals, audience, voice, success criteria — into the ICM intake. From day one, the work has a structure your auditors recognize.

    • · Time-on-task baseline
    • · Repetitive surface inventory
    • · brief.md · voice.md · _PROMPT.md
  2. 02

    Build & review

    ICM · 02 RESEARCHER · 03 PRODUCER · 04 REVIEWER

    We co-build the assets, prompts, Claude Projects, sub-agents, and automations the employee actually needs. Each pipeline stage writes its own reviewable output. The REVIEWER stage is a written human checkpoint — the same audit your risk and legal teams would ask for, embedded in the workflow.

    • · Custom Claude Projects & sub-agents
    • · Agent & automation pipelines
    • · findings.md · draft-vN.md · audit.md
  3. 03

    Ship & multiply

    ICM · 05 SHIPPER · packet.json

    We export a portable packet — Markdown plus the manifest, zipped — that your employee uses on Monday and their teammates inherit on Tuesday. A 90-day roadmap, weekly check-ins, and a productivity dashboard turn one packet into a department's operating system.

    • · Portable packet handoff
    • · Weekly throughput review
    • · Quarterly executive ROI report

04 — Deliverables

Not a certificate.
A packet you own.

Every engagement ships a portable ICM packet — Markdown plus a manifest, zipped — written to your storage. Below is what's inside. All of it readable by humans, replayable by any AI.

i.

Portable Packet

.zip · Markdown + manifest

The whole project — every stage, every artifact, every review — bundled into a single zipped folder any AI agent can pick up and continue. Vendor-neutral. Yours.
ii.

Audit Trail

04_review/audit.md

A written human-review checkpoint at the REVIEWER stage — the same record your risk, legal, and compliance teams would ask for, produced in the natural course of the work.
iii.

Prompt Library

01_frame/_PROMPT.md

A versioned set of prompts and Claude Projects tuned to the employee's recurring work — written as ICM stage briefs, diffable like code.
iv.

Agent Pipelines

03_produce/assets/

Multi-step agents wired to Slack, Notion, Drive, Salesforce — whatever they live in. Each step a named stage that can be re-run, replaced, or migrated.
v.

Playbook & Roadmap

05_ship/final.md

A written SOP describing how the role now operates — plus a sequenced 90-day roadmap of what to automate next, when, and why.
vi.

Productivity Dashboard

packet.json telemetry

Hours-saved, output-quality, and adoption telemetry the leader can quote at the board — derived from the same packet your auditors review.

05 — Tooling Philosophy

Claude-first.
Method-portable.

We've made a deliberate bet on Anthropic Claude as today's best executor of an ICM pipeline — for its reasoning quality, safety posture, agent ergonomics, and the maturity of the surrounding ecosystem (Projects, MCP, Claude Code, Computer Use). Where a different tool wins for a specific job, your employee learns that too.

But the method is platform-agnostic. ICM packets are Markdown plus a manifest — readable by any AI you adopt in 2027 or beyond. When the frontier moves, your work moves with it. No re-platforming. No re-training. No vendor holding the keys.

The result: fluency, not fanaticism. A workforce that picks the right AI for the job — and a body of work that outlives whichever AI wins.

Anthropic Claude
Primary platform.

Projects, sub-agents, MCP, computer use, code execution — taught at the level your specialist uses daily.

ChatGPT & Gemini
Where they fit.

Tool-agnostic where it matters. Your team learns the right model for the right job.

Cursor & Claude Code
Engineering-grade.

Engineers leave shipping at 2–4× — with reviewable, idiomatic code, not slop.

Office stack agents
Where work lives.

Slack, Notion, Drive, Excel, Salesforce — wired together so AI happens inside the tools they already use.

06 — Free Hour

One employee. One hour.
One packet you keep.

Choose any employee on your team. We'll pair them with a certified specialist for a single focused 60-minute session — workflow framed as an ICM pipeline, first packet shipped, productivity baseline captured. You'll see, on a real human in your org, exactly what audited AI work looks like.

·
One real recurring task, framed as an ICM pipeline
·
One complete portable packet shipped — yours to keep, runnable on any AI
·
Written summary & ROI projection delivered to the requesting leader
·
Zero obligation. We earn the next hour.

Reserve a free hour

By submitting you agree to be contacted by an Atlas specialist. We don't share your information.

07 — For Leaders

You don't need another platform.
You need a method your people, your auditors, and your next AI vendor all agree on.

Atlas Tradecraft is the layer between the AI tools your CIO selected and the workforce who needs to wield them — taught through ICM, so the work is productive, the quality is defensible, and the artifacts are portable. Dormant licenses become multiplying throughput. Every output is reviewable. And the playbook propagates across departments without another vendor cycle.

i.
0×

Productivity

Average task throughput

Across knowledge-worker engagements at the 30-day mark — measured per ICM pipeline.

ii.
0 hrs

Quality

per employee, per week — reclaimed and auditable

Hours saved on output that now carries a written human-review trail.

iii.
0×

ROI

Engagement ROI in year one

Fully-loaded labor savings + accelerated output + avoided rework once REVIEWER gates catch errors upstream.

Reference figures from pilot cohorts. Your engagement is measured against your own baselines.

08 — Specialists

Senior operators.
ICM-certified. Not a content team.

Every Atlas specialist has shipped real production work with Claude and the modern AI stack — and certifies against the ICM curriculum before stepping into an engagement. Each specialist recertifies on every major model release, so the method stays current as the frontier moves.

i.

Engineering Track

For staff engineers, platform teams, SREs.

Claude Code · Cursor · MCP · Agent SDK. ICM-structured production code, reviewable like prose.
ii.

Knowledge-Worker Track

For analysts, marketers, ops, finance, legal.

Projects · Prompt design · Document agents. Recurring role artifacts, rebuilt as ICM pipelines.
iii.

Sales & CX Track

For AEs, CSMs, support leaders.

Pipeline agents · Call summarization · CRM rituals. AI inside the deals you're already running — every output packet-auditable.
iv.

Executive Track

For C-suite + their EAs.

Strategic AI literacy + ICM fluency. 90-minute private sessions for the leaders deciding the policy.

A note from the field

“We'd written off our AI rollout as a sunk cost. Atlas spent a week with one analyst — by month two her whole team was clearing 30% more deals, and our risk committee finally had something to audit. We've now scoped the program across four divisions.”
Chief Operating Officer Fortune 500 Financial Services

Common Questions

Anticipating the obvious ones.

What is the ICM Method? +
The Interpretable Context Methodology (ICM) is an AI-native framework where the folder structure itself is the agent. One brief in; a numbered, auditable pipeline out — FRAMER → RESEARCHER → PRODUCER → REVIEWER → SHIPPER — each stage producing reviewable artifacts. The output ships as a portable Markdown + ZIP packet any AI can continue. The tool that implements ICM lives at icmproject.com; Atlas teaches it to your workforce.
How is ICM auditable? +
Every stage of the pipeline writes its inputs, prompts, and outputs to a numbered folder. The REVIEWER stage is an explicit human checkpoint with a written audit. Your risk, legal, and compliance teams can read the whole production trail in plain Markdown — no proprietary log formats, no vendor-only telemetry.
What does portable mean — and why does it matter? +
Every project exports as a packet: Markdown files plus a packet.json manifest, zipped. Any AI agent — Claude, ChatGPT, Gemini, an internal model, whatever you adopt in 2027 — can pick up that packet and continue the work without context loss. No vendor lock-in. The artifacts outlive the engagement and outlive the model.
How is this different from the AI training our vendor already runs? +
Vendor trainings teach features. We teach a method. Atlas embeds a certified specialist one-on-one with an employee, rebuilds their workflow as an ICM pipeline, and hands back a portable packet your auditors can read and any AI can continue. It's the difference between a cooking class and a kitchen system you keep.
Why Anthropic Claude as the primary platform? +
Reasoning quality, the safety and reliability profile enterprises require, and the maturity of the agent ecosystem (Projects, MCP, sub-agents, Claude Code, Computer Use). Claude is today's best executor of an ICM pipeline. But the method is platform-agnostic — packets keep working when models change.
How do you handle data security & IP? +
All engagements run inside your tenancy and your contracts. Specialists sign NDAs by default and operate under your DPA. We can run fully air-gapped where required. ICM packets are your IP — written to your storage, owned by you.
What's the catch with the free hour? +
There isn't one. We're confident enough in the outcome of one focused hour — one employee, one ICM pipeline, one packet shipped — that we'd rather invest it than pitch a deck. If we don't earn the next conversation, that's on us.
Can we scope a department-wide pilot? +
Yes — most engagements graduate from a single employee, to a job function, to a department. Because every output is a portable packet, what one employee builds is immediately reusable by their teammates. We provide an executive ROI dashboard so the case for expansion writes itself.

One brief in.
One packet out.

Pick one employee. Give us one hour. We'll ship one ICM pipeline — auditable, portable, runnable on any AI — and show your leadership team, on a real human in your real org, what high-quality AI work actually looks like.

Claim a free 1-hour session Try the method at icmproject.com Talk to a partner