AI Software Development
Generative AI consulting
Where generative models earn their place in a workflow, and where they do not
- Duration
- 3 to 5 weeks
- Ladder stage
- Proof
- NATIVE stages
- Target, Integrate
The situation
When this engagement is the right one
Generative models are good at drafting, summarising, classifying, and transforming. They are poor at arithmetic you cannot check and at decisions with no reviewer. Most disappointment traces to using them for the second category.
You are probably seeing
- Enthusiasm is high and no workflow has changed.
- Output quality varies and no one has defined what good looks like.
- Every function is running its own experiment with a different tool.
What we do
The work, in the order it happens
- 01Classify candidate tasks by tolerance for error and cost of review.
- 02Build one reference implementation with a quality bar and reviewer.
- 03Set the workforce prompt standard using the RAO framework.
- 04Define what generative models will not be used for, in writing.
What you get, and keep
- A task suitability map with the discard list included.
- One reference implementation with a documented quality bar.
- The RAO workforce prompt standard.
- A written exclusion list.
Prerequisites
- A function willing to have its work observed and measured.
Not included
- We do not run generic prompt-writing training with no workflow attached.
Where this sits in the method
Commercially this is a Proof engagement on the Proof, Product, Platform ladder.
FAQ
Questions we get asked
- Is this different from LLM consulting?
- This is the decision layer: which tasks and which controls. LLM application engineering is the build layer: models, prompts, evals, and cost.
Related engagements
Work that usually sits either side of this
AI Software Development
LLM Application Engineering
Model selection, prompt architecture, eval harnesses, observability, and token-cost control
AI Strategy and Operating Model
Board and Executive Briefings
Half-day and full-day sessions for boards and leadership teams, built on the NATIVE framework and the RAO prompt standard
Next step
Ready to scope generative ai consulting?
Bring the pilots you already have running. The first call is a scoping conversation, not a pitch.