Myelin · AI training and upskilling programme

Teaching your teams to make the most of AI at work.

They already use these tools. Few can tell a sound answer from a confident wrong one, because nothing in the output marks the difference. Myelin builds that judgement, in person, at three levels.

Mitochondria builds agentic AI systems for enterprises. We teach this because we build it.

Launching soon
I What is this resting on? Commission II What is not here? Omission III What would I have said? Anchoring IV What am I carrying? Accountability AN ANSWER ARRIVES A DECISION THAT HOLDS
Five competences, three levels
Written by the team that builds the systems
In person across the EU, UK, Australia and India
No client data, no system access
The problem

Wrong answers arrive looking like right ones.

The cost surfaces later. A figure that was invented, guidance that had gone stale, a decision nobody can account for when a client asks.

Seniority is no defence

Automation complacency, the habit of checking a system less often than the work requires, appears in expert and inexperienced people alike. It grows under competing demands on attention, which is to say it grows on a busy day. Awareness of it changes little.

Parasuraman and Manzey, Complacency and Bias in Human Use of Automation, Human Factors 52(3), 2010.

Two errors, not one

Machine advice produces errors of commission, where someone acts on a recommendation that was wrong, and errors of omission, where someone misses what the system never raised. The second is quieter and harder to catch, since nothing in the work shows that anything was missed.

Mosier and Skitka, 1996; Parasuraman and Riley, Humans and Automation, Human Factors 39(2), 1997.

So we build the sessions around practice. People work through answers that look sound and are not, until checking becomes reflex.

The curriculum

Five competences, taken further at each level.

What a participant can do afterwards, and can be seen doing.

OneLocate it
Find where these systems already sit in the work, including the parts that arrived quietly: summarising in a mail client, drafting in a document editor, ranking and scoring inside software the organisation has run for years.
TwoExplain it
Give a plain account of how an answer was produced, what the system could see, and what it had no way of knowing. Enough to hold a conversation with a supplier or a colleague on level terms.
ThreeJudge it
Recognise the failures that read as correct: invented detail that fits the surrounding text, confident claims resting on stale information, summaries that drop the one qualification that mattered, references to sources never consulted. Then apply the Check under time pressure.
FourFeed it
Decide what may go into a system and what may not. Client material, personal data, contracts, anything covered by an agreement with somebody else. What changes between a personal account and an approved tool.
FiveDeclare it
Record what a system contributed to a decision, and answer a colleague, a customer or an auditor who asks whether AI was involved. Includes raising a concern without it costing the person who raises it.
The method

The Myelin Check

Four questions, asked of any output before you act on it.

What is this resting on?
Against errors of commission

Find the source. Check the claims appear in it, and that it is current. A confident sentence with nothing beneath it is the common failure here.

What is not here?
Against errors of omission

Ask what a good colleague would have raised. Systems answer the question put to them, so the risk sits in the one nobody asked.

What would I have said?
Against anchoring

Set the answer aside, reach your own view, then compare. Reading first and judging afterwards leaves one answer on the table.

What am I carrying?
Against diffuse accountability

The system logs what it did. Only a person can say why they accepted it, and that is the part anyone will ask for later.

Why Myelin

In the body, myelin is the sheath that wraps a nerve fibre. It insulates the path and keeps the signal consistent, so what travels along it arrives faster and intact, and it grows stronger with repeated use.

Myelin does the same for judgement at work. The sessions layer method and technique onto the knowledge your teams hold, allowing system capabilities to be efficiently extracted, reducing business risks.

Method

The sessions are built to stick.

Practice over explanation

Teaching stays focused. The Check is applied to prepared outputs, considering the context of everyday work constraints.

Parasuraman and Manzey, 2010
Meet a wrong answer here

Better in the room than in front of a client. Exercises use work that reads well on a first pass and falls apart on a second.

Bahner, Hüper and Manzey, 2008
Something to take back

Everyone leaves with the Check and the worked examples from their own function. Optional quarterly check-ins with us.

Baldwin and Ford, Personnel Psychology 41, 1988
Three levels

Choose a level by what people do with the answer.

Using it, judging it, or answering for it. We sort this with you on the scoping call, and most organisations run two.

I
Foundations

For everyone working with these tools day to day. All five competences at working depth, and the Check practised until it runs without prompting. Participants leave able to say what the tools on their desk decide and what they leave to a person.

II
Practice

For those whose work turns on evaluating what a system produces. Longer exercises, harder material, the failure patterns in detail. Participants leave able to defend a decision to accept or reject an answer.

III
Governance

For those who set policy and answer for it. Disclosure, oversight, and the questions worth asking before a system is approved. Closes with a written position the organisation can stand behind.

ONE CURRICULUM I II III LOCATE IT EXPLAIN IT JUDGE IT FEED IT DECLARE IT

Every competence in every room, taken further at each level.

Format

Simple to book, simple to run.

A scoping call, a date, a room. Sessions run in person on a fixed curriculum, so there is nothing to design and nothing to approve. Exercises come from the work itself: a summary drafted for a client, a supplier's reply, a set of figures, a policy question. We write them by function, so the strategy room and the sales room handle different material, and we keep them current as the tools and capabilities evolve.

If your sector or your own processes would suit material shaped around them, we are glad to build that. Mention it on the call and we will scope it separately.

Facilitators are trained on the same curriculum and the same exercises, so the session holds wherever it runs.
Run in the working language of the room.
No client data required, and no access to your systems.
A record of attendance and exercises completed, for each person and for the organisation.
One fee, quoted up front, covering scoping and delivery.

That record also supports ISO/IEC 42001 clause 7.2, should your organisation be working toward the standard.

Sushrut Munje, co-founder and lead facilitator

Why the material stays current

We build on these platforms and see new capabilities early, so the curriculum keeps pace with what lands on your desk.

NVIDIA Inception program
Member
OpenAI Select Partner
Partner Network
Google for Startups
Cloud Program
Microsoft for Startups
Member

Mitochondria is a member of the NVIDIA Inception program and of the OpenAI Partner Network, and participates in Microsoft for Startups, Google for Startups Cloud, Claude for Startups and the Sarvam Startup Program.

The first cohorts are being scheduled.

We have been building this for a while and are looking forward to the first rooms. Dates, sizes and fees follow shortly.

Launching soon