Skip to content

Field support · organizations

Real work first. AI in the gesture.

For a leadership team, a prevention team or an SME: observe the activity — DUERP (mandatory occupational risk assessment), MSDs, tools, agents — and decide what to frame, integrate or refuse before investing.

Email Julien

Quote-based — scoped before you invest.

Activity ergonomist since 2020 — field work, not AI consulting. AI lead for CPME Réunion (employers’ union).

Diagnosis

What wears work down.

A compliant DUERP can still be unusable. Activity diagnosis connects the signals — DUERP, MSDs / psychosocial risks, QVCT (quality of work life), absenteeism — to the situations where risk actually appears, so the first change is defensible.

  • Start from the field signal, not the document alone
  • Work situations: gestures, pace, tools, compensations
  • Prioritized actions — not a report that gathers dust
diagnosis/activity
0%
❯ Activity diagnosis — logistics · 12 operators.
Field observation...
▸ DUERP (mandatory occupational risk assessment) signal: shoulder MSDs — 3 cases / 12 operators
▸ Work situation: handling, 2h rotation, 4 stations
Observe gestures and postures field [done]
Operator interviews field [running]
Observe constraints and compensations
◆ Verbatim: "I compensate with my back"
▸ Observable cause: pallet height cannot be adjusted
List disruptions, pace and cooperation field [done]
▸ Friction: AI already there — emails rewritten, no frame
Risk reframed as a discussable cause
◆ Prioritize: pallet station > rotation > training
Prepare actionable recommendations field [running]
✓ Priority 1: height adjustment + station rotation
Output: observable activity finding — not a report that gathers dust
▸ Operator 4 interview: "we wait for the machine"
▸ Station 3 rotation: 2h with no real break
▸ QVCT (quality of work life): mental load across multiple screens
Cross DUERP, MSDs and observed situations
◆ Field synthesis in progress...
Recommendations: station, pace, tool — not AI first
▸ Next pass: validate with prevention team

Situated AI

What AI changes in the activity.

Same field file: observe where AI already acts, decide responsibility and supervision, measure what changes in the loop — without stacking tools or promises.

  • Start from real activity, not from the tool
  • Connect agent → responsibility → supervision
  • Decide what gets delegated and what stays under human control
logistics · 12 operators — same diagnosis thread

▸ supplier emails rewritten with AI, no frame
▸ double entry: ERP + field sheet
▸ prescribed vs real task at pallet station
▸ attention: customer follow-ups at end of rotation

output: real use visible — not the tool demo

Training

Build skill before investing.

Individuals, managers and teams: understand what AI moves in cognition, attention and skill — so they can decide what to delegate before buying a tool. Keep the skills while you delegate — not one more training.

  • Start from real gestures and decisions, not from slides
  • Connect usage, supervision and responsibility
  • Decide before investing — not after
HR wants ChatGPT training for our 12 logistics operators. Where do we start?
From real work: emails, pallets, rotation — not the tool demo. Before buying anything, decide what to refuse.
Emails are faster, but operators verify less. What is the risk?
Vigilance atrophy — delegation, checkpoints, responsibility before anything is sent.
We are not looking for one more generic training.
Then use cases from the field file — to decide, not to tick a box.

In the field

DUERP (mandatory occupational risk assessment), MSDs, mental load, tools — the signals the activity makes legible.

  • DUERP (risk assessment) & prevention

    Make the risk document serve real workstations — not compliance alone.

  • MSDs / psychosocial risks

    Read the constraint at the workstation, not only the symptom.

  • QVCT (quality of work life) & mental load

    Work situations: pace, disruptions, multiple screens.

  • Absenteeism

    Look for the cause in the activity, not in the HR statistics.

  • Tools & process

    Double entry, ERP friction, AI already present without a frame.

  • AI or agent project

    Integrate it into the work gesture — supervised, not stacked on top.

  • Work situations

    Gestures, workstations, rotations, observable compensations.

  • Cognition & attention

    What AI moves in the gesture — skill, vigilance, errors.

  • Supervision & responsibility

    Who validates, who takes back control, what gets refused.

Method

Observe · Decide · Measure.

One loop before investing — the same field discipline.

01

Observe

Real work in the field: gestures, workstations, where AI is already entering without a decision. Make the activity observable — not one more report.

02

Decide

Delegate, keep human, refuse. Nothing leaves without validation — responsibility and supervision are written down.

03

Measure

Cognition, attention, skill, errors — on the activity, not on the demo. Keep it if the work changes; stop it if it does not.

FAQ

Common questions.

What is an activity ergonomist?

An activity ergonomist analyses real work — what people actually do, not what the job description says. Julien Talbot is an activity ergonomist based in Réunion (France): work diagnosis (DUERP — mandatory occupational risk assessment, MSDs, psychosocial risks, QVCT — quality of work life), situated AI integration and team training.

Do you work outside Réunion?

Yes. Julien Talbot is based in Réunion and works across the island in person. Missions in mainland France or remotely are scoped case by case, combining video calls with grouped observation time.

How does an engagement start?

Three lines by email are enough: the situation, the decision at stake, the visible constraint. A first conversation picks the right door — diagnosis, situated AI or training — before any commitment.

How is this different from a classic AI consultant?

An AI consultant starts from the tool; an activity ergonomist starts from the work. Julien Talbot observes the activity before recommending what to frame, integrate or refuse, and sells no licences or development — the recommendation stays independent of the solution.

Read

Field cases, in writing

Work together

Three lines are enough.

The situation or case, the decision to make, the constraint already visible. A first exchange to choose the right entry point — diagnosis, situated AI or training.

or by email: contact@julientalbot.com

Rather an entrepreneur? Hire an agent →