IWise / Method

Four phases. Four things you keep.

Any firm can list discovery, build and deploy. The part worth reading is what you receive at the end of each phase, and the fact that the next one does not open until it exists.

01

Prove it against your real data.

A paid technical discovery. We take one workflow, connect it to the systems that actually hold the answer, and run it on your records. Not a demo, and not a slide. If it will not work we would rather find out here, and so would you.

You receive

A written architecture, and a working thin slice running against your own data.

02

Build the agent set, and the way you know it is wrong.

The agents, the MCP servers into each of your systems, the guardrails they cannot cross, and the graded eval set that catches failures before your customers do. The eval set is not optional; it is the thing that makes the system maintainable after we leave.

You receive

The system, the eval set and the guardrail spec, in your repository.

03

Ship it behind your access controls.

Into production inside your boundary, with traces, cost and latency visible from the first day it runs. Failures surface to a named person, not to a dashboard nobody opens.

You receive

A runbook, an escalation path, and a written on-call arrangement.

04

Compound, or hand it over cleanly.

Models change underneath production systems without asking. Every quarter the graded set runs again and the deltas get written down, whether they are flattering or not. Keep us operating it, or take it in-house with the runbook.

You receive

A quarterly evaluation against the graded set, with every regression recorded.

Two people working through a plan at a whiteboard.
Phase 01 is a paid discovery. The architecture gets written down.
Working together

Three shapes. None of them per seat.

We price the engagement, not the person. There is no seat count and no token allowance.

Two colleagues working through a plan at a whiteboard.
Nothing moves to the next phase until the artefact from this one exists.
Fixed price

Technical discovery

One workflow, connected to the systems that hold the answer, running on your data.

  • A written architecture
  • A working thin slice
  • A go / no-go recommendation
Fixed scope

Implementation

The full agent set, the integrations, the guardrails and the eval harness, shipped into production.

  • The system, in your repository
  • The graded eval set
  • A runbook and an escalation path
Ongoing

Build and operate

We keep running it. Quarterly re-evaluation, model changes absorbed, new workflows added.

  • Quarterly evaluation against the graded set
  • Named on-call
  • Clean exit with the runbook
Questions

The ones worth asking.

What happens when the model changes underneath us?
Every system we run carries a graded eval set, and that set reruns whenever a provider ships a change. The deltas get written down whether they are flattering or not.
What if it breaks at 2am?
The runbook that answers this ships at the end of phase 03, before anything goes live.
What do you need from our team?
Two people. A product owner who can say what a correct answer looks like, and a systems owner who can grant access and sign off on what the agent is allowed to write. If neither exists yet, that is worth solving before you spend money with us.
How is this different from hiring someone in-house?
A hire is one person, six months out, with no eval harness and no second opinion at 2am. You get a pod that has shipped this before, and a written methodology with gates you can hold us to. When it is running steadily, you can take it in-house with the runbook.
Do you also do placement or offshoring?
IWise runs a placement practice. It is a separate line of business and it is not what this site is about. Ask us if you need it.

Start at phase 01.

A paid technical discovery on one workflow. You leave with a written architecture and a working thin slice, whether or not you continue.

Start a discovery See the five systems