Before you start: an approved solution set from Define, with an owner for each change. See Define your solution set.
The delivery loop
1
Simplify the work
Remove waste and reengineer what’s left before you automate anything. Make the human handoffs explicit, and start the system, data and organizational changes the outcome depends on.
2
Design, build and release
Design the change, build it, and test it on representative work. Release only when quality, controls, acceptance and operation all pass. If they don’t, hold the release, fix what’s blocking it and run the acceptance tests again.
3
Run it in production and adopt it
The change is live when eligible real work succeeds through it, again and again, with a named operator and a support route. People use it, inspect what it did and recover when it goes wrong.
4
Measure the value
Recalculate against the baseline from Discover, using comparable cases and the full cost of running the change. The first iteration is complete when the customer accepts the measured benefit.
5
Decide what changes next
Sustain it, update it, build the next one, re-agree a changed objective, scale it, or retire it. Your Work Brain keeps learning from the work, so the next decision starts from current evidence.
How each solution type runs through it
Within’s part in each type, and what your teams own, are in Define your solution set.
Two ways to build automation
Business users: skills and simple agents
Build in Advisor, or in a chat app connected to the Within MCP. One bounded task at a time, reviewed before it’s shared.
Developers: compound agents
Build in the coding agent your team already uses, with the Within ADK guiding the sequence from objective to deployment.
In this section
Redesign the process
Remove waste and reengineer, then pilot, document and roll out the new way of working.
Using the Within MCP
Connect Within to your own AI client, and how it compares to Advisor.
Build an agent with the Within ADK
Specify, build and validate an agent from your observed workflows.
Validate an agent before release
Test against ordinary, edge and failure cases, then pilot in shadow and assisted mode.
Run in production and adopt
What has to be true before people can rely on it in real work.
Maintain and improve agents
Keep an agent current as the work changes.
Measure the value
Recalculate against your baseline so the value claim holds up.
Need help? Use the in-app chat — click the chat bubble in the bottom-right corner of Within (staffed 24×5) — or email support@within.ai.