Chat without a harness
Useful for thinking. Thin when the job is a deliverable from files, or a project you will run again.
Helix Craft · Claude
Get real value from Claude Chat, Cowork and Code, on the work you already do.
You get real value from Claude by matching the harness to the work. Use Chat to think, Cowork to deliver a defined piece of work from your files, and Code for a project you run and maintain. Helix Craft trains teams to make that choice on their own work, with a person still deciding.
Practice Craft
Guided practice on real work, for leaders and teams using Claude.
Helix Craft is built on Claude, and its key leaders are trained on it — the same Chat, Cowork and Code your team will use.
The stall
Why Claude stalls at the chat box.
Most leaders now have AI, but few get past the chat box. People ask questions; they stop short of connecting Claude to their own documents and drives, or building it into how the team works. Models sit at fingertips. The method is the missing layer.
How much help should this task get?
The hard question is agency: Assist, Augment or Adapt — and what must stay yours. AI checks; a person decides.
The choice
Three ways Claude works, and how to choose.
The model is the intelligence. Chat, Cowork and Code are three ways of equipping a model to work. Name the stop point before you start. “Draft for my review” is a different job from “update the working files”. More autonomy is useful only when the work becomes easier to verify.
Chat · Assist
Helps me think; I act. At the helm.
Cowork · Augment
Works through several steps; I review. In the loop.
Code · Adapt
Works directly on files inside rules; I remain on call.
As of October 2026, Anthropic is bringing Chat and Cowork together into one Claude, rolling out gradually by plan. Name the workflow you need — think, deliver or build — rather than looking for a particular button.
Chat · Assist
Turn a messy stakeholder request into a clear ask.
My role: At the helm
What I would check: Does the ask name the outcome, the constraints and the stop point?
Chat · Assist
Stress-test a draft options paper with counter-arguments.
My role: At the helm
What I would check: Which claims need evidence before a person signs them?
Cowork · Augment
Synthesise interview notes from a folder into a structured report.
My role: In the loop
What I would check: Are sources visible, exclusions respected and the tone right for the audience?
Cowork · Augment
Organise a project folder into a clean, agreed structure.
My role: In the loop
What I would check: Was anything removed without approval, and does naming match the team standard?
Code · Adapt
Build a recurring report pipeline you will run every week.
My role: On call
What I would check: Do acceptance examples pass — ordinary, incomplete and reject cases?
Code · Adapt
Maintain a small internal tool that updates working files under rules.
My role: On call
What I would check: Is there a review gate before changes land, and do the rules still hold?
Chat
Claude Chat: help me think.
Use Chat for briefings, critique, options and turning an unclear request into a clear ask. Connect it to sources with connectors and projects — the Connections layer — so answers sit on your material, not only the public web.
Assist
AI helps me think, and I decide. Chat can draft documents and use connectors, but it doesn't work directly in your folders the way Cowork does. The difference is the shape of the assignment.
When Chat is enough
You need a sharper question, a challenge to a draft, or options before anyone touches the working files.
Cowork
Claude Cowork: what it is, and how to hand over a deliverable.
Cowork brings agentic capability to knowledge work beyond coding. It takes on multi-step tasks, works with local files on desktop, can coordinate sub-agents, and supports projects and scheduled tasks on paid plans. You stay in the loop.
Name the assignment
Name the source folder, the result and the exclusions. Example shape: “From the interview-notes folder, produce a two-page synthesis for the steering group. Exclude raw transcripts. Flag anything you are unsure about.”
Permission modes, in plain English
As of October 2026, Manual is the default: Claude asks before it acts. Auto keeps working while Claude checks each action for safety and blocks anything it judges unsafe. Some plans also show Skip, which checks nothing; Anthropic says to use it only when you completely trust every action, connector and file involved. Permanently deleting files needs your explicit permission. Anthropic notes Cowork has unique risks because it is agentic and has internet access, so stay on Manual for consequential work.
Augment
AI carries several steps, and I review. What evidence would let me accept the result?
Scheduled tasks and projects
Only once the task is proven. Ask first whether the recurring work should exist at all — No Spoon before automation.
Code
Claude Code for knowledge workers: build something you'll run again.
Claude Code reads a codebase, edits files and runs commands. It is available in the terminal, desktop app, browser, VS Code and JetBrains. For knowledge workers, that means repeatable projects you can review — not only software teams.
Write acceptance examples first
Ordinary case, incomplete case, reject case. Arrange competent review before Code touches important files.
Choose the right harness
If it is a one-off deliverable from files, use Cowork. If the task is unclear, use Chat. Code earns its place when you will run and maintain the result.
Adapt
AI works on files inside rules I set, and I stay on call. Optional, and earned — the Craft Super Licence path reflects work you have done, not videos you have watched.
When Helix Studio builds
When the work has earned a commissioned build, Helix AI Studio can take idea to prototype in four weeks.
Sacred Timeline
Protect the timeline before Claude touches important work.
Before AI changes important work, protect the timeline: captures, experiments, proposals, then human review. Admin settings for Team and Enterprise — web search on or off, Auto mode availability — stay with your organisation’s admin.
Captures and experiments
Keep a record of what Claude saw and tried, so review has a trail.
Proposals before write
Consequential changes land as proposals a person accepts, not silent edits to the Sacred Timeline.
Adoption
From one champion cell to the whole team.
Leaders go first. Start with a small champion cell on real work, then grow team cells that share a review standard. Account-level skills and connectors follow the account across Chat, Cowork and Code; command-line additions stay on that machine.
The licence reflects practice
The Craft licence reflects work you have done, not videos you have watched. Claude Academy, Anthropic’s free course site, covers free generic courses; Practice Craft trains on your own work.
Shared team standard
Agree what “good enough to accept” means: stop points, review, and when Manual stays the default.
Both tools
Using Claude and Copilot together.
Many organisations have both. Pick the harness by the work, not the brand. Helix Craft language stays the same: Assist, Augment, Adapt — judgement kept.
Questions
FAQ
How do I get the most out of Claude?
Match the harness to the task. Use Chat to think, move one recurring task into Cowork, and keep the decision human. Helix Craft trains that choice on your own work.
What is the difference between Claude Chat, Cowork and Code?
Chat helps you think; Cowork delivers a defined piece of work from your files; Code builds a project you run and maintain. Agent Watch compares the three surfaces in more detail.
What is Claude Cowork used for?
Multi-step knowledge work from your files — reports, organised folders, synthesis. You review. Name the source folder, the result and the exclusions before you start.
Is Claude Cowork safe for work files?
It works in folders you connect, deletions need approval, and you choose the permission mode. Anthropic flags agentic risks. Use Manual for consequential work.
Do I need to be a developer to use Claude Code?
No. You can use Claude Code in the desktop app or a browser without a terminal. Use Code for repeatable projects you can review. Otherwise use Cowork.
What is Claude Code, and how do I use it well?
Claude Code is Anthropic’s agentic tool that reads files, edits them and runs commands, in the terminal, desktop app, browser and IDEs. Use it well by writing acceptance examples first — ordinary, incomplete and reject cases — and arranging review before it touches important files.
Is there training for Claude?
Yes. Claude Academy, from Anthropic, is free and generic. Practice Craft trains leaders and teams on their own work, with a person still deciding.
Can a whole team use Claude the same way?
Account-level skills and connectors carry across surfaces. Agree a team standard for review, stop points and when Manual stays the default.
Does Helix work outside Adelaide?
Adelaide, working across Australia and worldwide.
Where we work
Location.
Adelaide, working across Australia and worldwide.
Let’s find the starting point
Bring one real Claude task into Craft.
Open the free sampler and leave with a named task and the harness it needs. No account needed. Guided practice by invitation: conversation and fit first, numbers on the call. A Rhythm Briefing (45 minutes) is available as the executive-first conversation. For leaders deciding which AI work to fund, start with Clarity Scan.
Bring one real task: try Craft