Keep building when the project gets complicated.
A working demo doesn’t tell the next agent why you chose that architecture, which permission rules matter or what still needs testing. Engineering With AI keeps that context with the project, from the first idea through planning, delivery, review and the next change.
Install the harness directly from npm. Premium personas are optional.
“Add an export button” leaves a lot unsaid.
Which records? Who can download them? What must never be included? Intent Studio helps you develop a request into work that can be reviewed before an agent starts building.
Make the important details hard to miss.
In this fictional customer portal, an export isn’t useful if it leaks another customer’s records. It also needs to work from a keyboard and explain what happens when a download fails.
The draft names those outcomes. They can inform implementation and testing, and give the reviewer something more specific to check than “the feature works”.
These are genuine captures of the local harness using demonstration data. The draft is not approved, and the criteria shown are not passing test results.
The work doesn’t disappear between “plan” and “done”.
The deliver skill coordinates fourteen stages across intent, reconciliation, planning, validation, Build, testing, delivery and learning. Each stage leaves the next one a clearer starting point, while named people retain the decisions that need human authority.
See where the work is, without turning on every specialist workspace.
The current dashboard starts with a simpler operating model and a collapsible sidebar. Project, portfolio and governance views remain available when the work needs them; they don’t have to crowd every project from day one.
Start with the problem that’s hurting now.
You don’t need to adopt every capability on day one. Each route solves a recognisable part of the engineering problem, and they join up when the project needs more depth.
Understand the system you have
Map repositories, reconstruct missing knowledge, test the code against intended purpose and make uncertainty explicit before AI starts changing things.
Explore discovery and Archaeology → 02 / SHAPEShape the work before code
Turn a broad request into clear intent, architecture, impact, constraints and a delivery shape a person can actually approve.
Explore intent and architecture → 03 / DELIVERBuild, review and prove
Move through guarded delivery, independent review, standards, tests, security evidence, Manual QA and human acceptance.
Explore governed delivery → 04 / DESIGNDesign with people involved
Use real design systems, prototypes and relevant persona lenses to find misunderstanding while it’s still cheap to change.
Explore design and prototype review → 05 / REMEMBERKeep project memory
Make decisions, evidence, meetings, lessons, errors and open questions findable without feeding the model the whole repository every time.
Explore context and knowledge → 06 / SCALEScale across teams and projects
Coordinate portfolios, repositories, shared blueprints, policies, resources and consultancy rollouts while each project keeps its own evidence and authority.
Explore teams and portfolio scale →The decisions need to survive the chat.
A skill helps an agent tackle a task. A coding harness also needs to know which task is authorised, what it depends on, which standards apply and what evidence is still missing.
Engineering With AI combines project-local records, a delivery pipeline, a dashboard and agent workflows. You can inspect the work without reconstructing it from a completion message.
- Before the change
- Archaeology and the Source Map help establish what exists. Intent, architecture and impact analysis help define what should change.
- During delivery
- Phase state, dependencies and execution leases support coordinated work. Tests, review findings and Manual QA remain different kinds of evidence.
- After the session
- SPECS, Mind Palace search, meeting evidence and retrospectives help preserve the decisions and lessons the next task needs.
Built under the pressure of a real platform.
The harness helped us build Backstory.
That meant dealing with a large, long-lived system spanning governance, privacy, risk, suppliers, assurance, knowledge, work management and more. It meant inherited decisions, multiple repositories, thousands of tests, competing priorities and the very human problem of remembering why something was built months after the conversation moved on.
The line count isn’t the achievement. Keeping that much work understandable, reviewable and able to move forward is the point.
Try the harness in a project you can experiment with.
Use the repository when you want to inspect how the system works. Start in the docs when you need the adoption route, the operating detail or the boundaries behind a particular capability.
Current status: The harness is open to everyone. The installation guide covers setup, updates and optional premium personas.