Approach

How we work, where judgment stays, and what you can expect.

Good deployment depends on a clear workflow, defined responsibilities, and human judgment where it matters. Here is how we approach the work.

Working principles

How we think about AI.

PRINCIPLE 01

AI helps us move faster, not skip the thinking.

Models can help with sourcing, drafting, and synthesis. The judgment still has to hold up in the room, so a person reviews the output when it informs a real decision.

PRINCIPLE 02

Structure matters more than clever prompts.

A useful system depends on the quality of its information, workflow, and design. Prompts are only one part of it.

PRINCIPLE 03

The output has to survive a real decision.

The output has to be clear enough to review and useful enough to support the next decision. That is the standard we design toward.

How an engagement runs

A clear path from the work to a useful output.

We make each stage, handoff, and decision point visible—including where human review is required before the work moves forward.

And after launch

Launched isn't the same as used.

We involve the people who will use the solution while it is being designed, prepare them to use it, establish responsibilities, document the workflow, and review early feedback after launch. Adoption is a shared effort, so we work with your team to support regular use and identify what needs improvement.

What we mean by “Works”

A solution should do the intended job reliably, fit the workflow, be usable by the people involved, support an outcome that matters, and remain manageable over time.

Boundaries & human review

Clear boundaries. Human judgment where it matters.

Before deployment, we agree what the system may do, what requires human review, and what remains off-limits. The level of review depends on the information involved and the consequences of getting something wrong.

The system can act within explicitly agreed limits. A named person reviews before the next action is taken. The system is not permitted to take this action at all.
We explain what information the solution uses, where it goes, and what controls are available—before we deploy. We agree where human review is needed, based on the information involved and the consequences of error. We also document the system’s known limits.
Tools & models

Tools chosen around the work.

We evaluate tools and models against the workflow, the information involved, the available controls, and what your team can realistically manage. When sensitive information calls for a private or self-hosted approach, we explain the options and trade-offs. Model choice is one part of the implementation; the working solution is the point.

What you can expect

Plain communication, clear boundaries.

  • ✓ Plain-English updates, with unfamiliar terms explained.
  • ✓ A realistic view of what AI can and can't reliably do.
  • ✓ Clear scope and boundaries agreed up front.
  • ✓ Agreement on how we'll tell whether it's working.
  • ✓ If something is unclear, it is our responsibility to explain it better.

AI in the workflow

The model is one part of the system.

We evaluate leading AI providers and their platforms against the work they need to support. Then we design the workflow around the information involved, the tools already in use, the controls required, and the people responsible for reviewing the output.

01 — AI providersEvaluated against the work
OpenAI
Anthropic
Google
Microsoft
02 — Deployment around the workDesigned and operated by ontopraxis
Workflow design
Information access
Human review
Controls

Where useful, we can design agentic workflows that gather context, use approved tools, and operate within defined limits.

03 — Where the work happensYour existing environment
Email
Documents
Spreadsheets
Knowledge systems
Systems of record

Available connections depend on the selected platform, access, and security requirements. Logos do not imply endorsement, certification, or partnership.

See how this thinking shows up in real work.

Explore our AI work Start a conversation