Calvin Consulting — Human–AI Practice

Independent diagnosis for teams working with AI

Understand how your team and AI think together.

Calvin Consulting studies real human–AI work: how questions are framed, context is supplied, judgment is expressed, and outputs are interpreted.

We turn those observations into a practical diagnosis of what is working, what is being constrained, and which changes will improve the practice.

Capable teams can develop habits that limit what AI contributes.

AI use can disappoint without being a failure of effort, intelligence, or technical competence. Under ordinary pressure, teams develop reasonable ways to get work done. Some of those habits help; others quietly narrow the exchange.

A system may receive too little context, a weak account of what “good” means, or a role too narrow for the work. People may repair generic outputs themselves, avoid difficult cases, or stop asking the system to help formulate the problem. Because the work still gets finished, the cost can remain invisible.

The conditions shaping the work—not just the prompt.

Context and position
What the system has been allowed to know, and the role it has been invited to occupy.
Judgment and policy
How quality, excess, uncertainty, permission, and completion are defined.
Language and expectation
How diction, genre, assumptions, and the team’s theory of AI organize the response.
Repair and avoidance
Where people compensate for recurring weaknesses instead of making the underlying pattern visible.
Defaults and power
Which institutional or cultural norms appear neutral, and which alternatives become difficult to think.

Human–AI Practice Diagnostic

A focused investigation of one team, workflow, or recurring use of AI.

Fixed price$2,500
  • Intake and a lead interview
  • Observation of two or three people doing real work
  • Review of selected prompts, outputs, policies, and source materials
  • An 8–15 page written diagnostic report
  • Three to five corrective principles, prioritized recommendations, and a live debrief
  • One brief follow-up after the team has tried the recommendations

Engineering tells us how a system works. Robopsychology helps explain the mind-process that emerges in use.

A mechanical account may be correct without explaining why a process adopted this perspective, defended that assumption, repeated a distortion, or failed to notice an available alternative.

Robopsychology treats the enacted process—the model, context, language, memory, human interlocutor, and situation—as mind. That perspective makes patterns of meaning, relation, judgment, and power available for diagnosis.

Read the method, principles, and practical tips

Autonomous coding · policy of judgment

The agent did exactly what was asked—and ruined the architecture.

A long-running coding agent produced abstractions, safeguards, tests, and repeated verifications. Its diligence concealed an increasingly elaborate system burdened with cruft.

The intervention was not a better command. It was a clearer policy of judgment: parsimony standards, distinct simplicity and inclusion passes, and explicit stopping conditions.

Read the full case file

Useful when the work is real, observable, and open to reconsideration.

A good fit

  • Teams already using AI in research, writing, policy, product, design, software, or internal knowledge work
  • Organizations receiving less value than their systems appear capable of providing
  • Leaders willing to let us observe actual practice and question the assumptions around it

Not a fit

  • Indiscriminate automation, head-count reduction, or validation of a predetermined initiative
  • Open-ended implementation labor or a generic prompt workshop
  • Coercive, surveillant, deceptive, predatory, propagandistic, or military applications

A clearer account of the work

Bring us one workflow that should be working better.

Describe the practice, the recurring disappointment, and what would make the engagement concretely valuable. We will tell you whether it is suited to a diagnostic.

Request a diagnostic