Table of Contents
Train engineers in AI by role, not with one course for everyone. Operators and managers need AI literacy: what a model or assistant can and cannot do, and how to check its output. Process and maintenance engineers need to frame problems and prepare data. A small group needs modelling and deployment skills, learned on your own plant data.
Key takeaways
- The need is broad. Employers in the WEF's 2025 survey expect 59 of every 100 workers to need training by 2030, and 85% plan to prioritise upskilling (WEF).
- Most engineers need judgement more than data science. In a field experiment, consultants using GPT-4 on a task outside the model's capabilities were 19 percentage points less likely to reach a correct solution than those without it (Dell'Acqua et al., 2023).
- Security agencies ask plant owners to train OT teams to interpret and validate AI outputs and to keep their manual skills (CISA et al., 2025).
- Training lags behind use: only 39% of people who use AI at work had received AI training from their company in Microsoft and LinkedIn's 2024 survey (Microsoft).
- In Algerian public tenders, Article 60 of Law 23-12 requires the specifications to provide, as applicable, for training and the transfer of know-how (Law 23-12). Use it to make vendors train your people.
Why train engineers in AI now?
Skill gaps are the barrier employers cite most. In the World Economic Forum's Future of Jobs Report 2025, based on a survey of over 1,000 employers, 63% named skill gaps as a major barrier to transformation, and AI and big data top the list of fastest-growing skills. Of every 100 workers, employers expect 29 to be upskilled in their current role and 19 to be retrained and redeployed by 2030. Companies headquartered in Central Asia and the Middle East and North Africa expect under 50% of their workforce to need training, against 67% in North America (WEF).
Your engineers are probably already using AI without training. In Microsoft and LinkedIn's 2024 survey of 31,000 people in 31 countries, 78% of AI users brought their own AI tools to work (Microsoft). In a plant, that can mean procedures, drawings or lab results pasted into a public chatbot. Training is the first control: people need to know which data may leave the site. Our guide to what AI vendors do with your data lists the questions to ask.
Algeria is building its own pipeline. In May 2026 the government reviewed a draft national AI strategy built on three pillars: data, digital infrastructure and human skills (Radio Algérie, May 2026). The National Higher School of Artificial Intelligence (ENSIA), created in 2021, graduated 105 students in June 2026 (Radio Algérie, June 2026). Graduates bring the methods but not your plant knowledge, and they will not reach every site. Most of the skills you need will come from training the engineers you already have. Our overview of Algeria's industrial digitisation gives the wider context.
Which AI skills does each role need?
Not everyone needs to code. Match the depth to what each person will do with AI.
| Role | What they need to be able to do | Depth | Typical format |
|---|---|---|---|
| Managers and sponsors | Choose problems worth solving, read a pilot result against a baseline, set rules for data and tools | Literacy | Short workshop on your own use cases |
| Operators and technicians | Use an assistant or a model output, check it against instruments and procedures, know when to ignore it and how to report errors | Literacy | Hands-on sessions on the actual tool, before go-live |
| Process, maintenance and quality engineers | Frame a problem as a prediction or a search, explain tags and events, judge model results, ask an LLM assistant precise questions | Applied | A course plus a project with a data team |
| Automation and OT engineers | Give safe data access to the DCS, SCADA and historian; understand where AI sits in the network | Applied | Project work with IT and the vendor |
| IT and security staff | Host, secure and update models and data on-premise; manage access | Applied | Vendor training and documentation |
| Future practitioners (a few engineers) | Clean and align plant data, build and test models, monitor them in service | Specialist | Months of courses and protected project time |
The last row is covered in our guide to building a first industrial data team. The other rows are most of your workforce.
What does AI literacy mean for engineers who will not build models?
Its core is knowing where the tool is reliable. In a pre-registered experiment with 758 Boston Consulting Group consultants, those using GPT-4 on tasks within the model's capabilities completed 12.2% more tasks, 25.1% faster, with more than 40% higher quality. On a task chosen to be outside those capabilities, they were 19 percentage points less likely to produce a correct solution than consultants without AI (Dell'Acqua et al., 2023). The authors call this a "jagged technological frontier": tasks that look equally difficult can fall on either side of it.
Plants face the same risk with higher stakes. The joint guidance on AI in operational technology led by CISA warns that OT staff may misinterpret AI outputs and lose manual skills through over-reliance. It recommends training OT teams on AI fundamentals so they can interpret and validate outputs, for example against other sensors or their own senses, and know what to do when outputs are invalid (CISA et al., 2025). The NIST AI Risk Management Framework likewise expects an organisation's personnel and partners to receive AI risk management training suited to their duties (NIST AI 100-1).
For a plant, a literacy module should cover:
- What a predictive model does. It learns from past data and can be wrong when the process moves outside that data.
- What an LLM assistant does. It writes fluent text that can be false, so answers should cite a source document that the user opens. See measuring hallucination in an industrial assistant.
- Which data may go where. What can be typed into a public tool, and what stays on the company's own systems.
- How to check an output against instruments, procedures and experience, and when to override it.
- How to report a wrong answer, so the people who own the model can fix it.
- How to work without the AI. The procedure to follow when it is unavailable.
How should engineers who will build models learn?
Start with free, well-known courses, then move to your own data quickly.
| Course | Level | Prerequisites stated by the provider | Language | Notes |
|---|---|---|---|---|
| Elements of AI, Introduction to AI (University of Helsinki and MinnaLearn) | Literacy | No programming or complicated mathematics | French version available | Free |
| Google Machine Learning Crash Course | Introductory machine learning | Variables, linear equations, graphs of functions, histograms, means; ideally some Python | English | Programming exercises in Python |
| fast.ai Practical Deep Learning | Applied deep learning | About a year of coding; high-school mathematics | English | Free; 9 lessons of about 90 minutes |
| Hugging Face LLM Course | LLMs and NLP | Good knowledge of Python; ideally an introductory deep learning course first | English, with a French translation | Free |
These courses teach methods on generic datasets. Plant data has its own problems: renamed tags, lab results stamped with the analysis time instead of the sampling time, shutdowns hidden in the historian. Engineers learn to handle those only on a real project. Give each trainee a problem from their own unit, a mentor (in-house or at a partner) and protected time.
How do you build an AI training programme, step by step?
- Start from the use cases. Train for the AI your plant will actually use in the next year or two. An AI readiness assessment identifies those uses and the gaps.
- Map roles to depth using the table above.
- Write the rules first. Decide which tools are approved and which data may leave the site.
- Train users before go-live, on the real tool. Operators learn more from their own procedures in the assistant than from slides. See operator assistants and digital work instructions.
- Run a practitioner track for a few engineers, combining courses with a project on their own data.
- Make the vendor teach. Put training and knowledge transfer in the contract, with named trainees and a handover test.
- Update the procedures. Add the AI tool, its checks and the fallback to the SOPs, and practise operating without it, as the CISA-led guidance recommends.
- Measure and refresh. Check whether behaviour changed on the job, and train again when the tool or the procedure changes.
What should a training clause in an AI contract say?
In Algeria, public tenders already have a legal basis for it. Article 60 of Law 23-12 on public procurement requires the contracting authority, as applicable, to provide in the specifications for "a mechanism to ensure training and the transfer of technology and know-how" related to the contract's object (our translation). Article 64 requires international tenders to set, for foreign bidders and subcontractors, a minimum threshold for employing national workers and qualified national staff, so that they can develop skills and gain experience (Law 23-12). Which buyers the law binds depends on their legal status. Our RFP guide summarises this; confirm with your procurement department and counsel.
Private companies can write the same into their contracts. A useful clause names:
- the people to be trained, by role, and the hours of hands-on training on your own infrastructure;
- documentation in the languages your teams use, covering the data pipeline, the model configuration and the operating procedures;
- a shadowing period in which your engineers run monitoring and retraining while the vendor watches;
- an acceptance test: your team retrains or redeploys a model, or updates the assistant's documents, without the vendor's help.
How do you know the training worked?
Attendance is not a result. The Kirkpatrick model, a common framework for evaluating training, has four levels (Kirkpatrick Partners):
| Level | Question | Plant example |
|---|---|---|
| 1. Reaction | Did participants find it relevant to their work? | Short survey after each session |
| 2. Learning | Did they gain the knowledge and skills? | Exercise: spot the wrong answers in a set of assistant outputs |
| 3. Behaviour | Do they apply it on the job? | Share of assistant answers checked against the source; errors reported; models reviewed by their owner on schedule |
| 4. Results | Did the intended outcomes follow? | The use case KPI, such as time to find a procedure or lab samples avoided |
Levels 3 and 4 take the most effort to measure, so decide how you will measure them before the course starts.
What we have seen at KABAS AI
Our two published projects are in formulation R&D, and in both, the client's people were part of the deployment. At ENAD Shymeca, the Virtual Lab implementation began with a discovery phase in which KABAS engineers worked with the client's R&D team on workflows and data structures, and included a training and adoption phase so that R&D staff could use the platform fully. Capturing knowledge before experienced researchers retired was one of the client's own concerns. At ADFC, structuring decades of formulation experience into a repository supports knowledge transfer across generations of technical staff.
The practical lesson is to train people on the tool they will use, with their own data and documents, while the project is running. With DALILE running on your own servers, engineers can practise on real procedures and reports without that material leaving the site.
Where should you start?
List the AI uses planned for the next two years and the roles that will touch them, then compare your current skills with the role table above. Our consulting team runs readiness assessments that cover data, infrastructure and team capabilities, and our data science team can work alongside the engineers you are training. Read more about KABAS AI, or contact us to discuss a training plan for your sites.
Sources
- World Economic Forum, The Future of Jobs Report 2025
- Dell'Acqua et al., Navigating the Jagged Technological Frontier: Field Experimental Evidence of the Effects of AI on Knowledge Worker Productivity and Quality, Harvard Business School Working Paper 24-013, 2023
- CISA, ASD's ACSC et al., Principles for the Secure Integration of Artificial Intelligence in Operational Technology, 3 December 2025
- NIST, Artificial Intelligence Risk Management Framework (AI RMF 1.0), NIST AI 100-1, 2023
- Microsoft and LinkedIn, 2024 Work Trend Index: AI at Work Is Here. Now Comes the Hard Part
- Law 23-12 of 5 August 2023 on public procurement, Journal officiel no. 51 (FAOLEX copy)
- Radio Algérie, Réunion du Gouvernement : la stratégie nationale d'intelligence artificielle à l'ordre du jour, 25 May 2026
- Radio Algérie, Baddari préside la cérémonie de sortie de deux promotions de l'ENSIA et de l'ENSM, June 2026
- University of Helsinki and MinnaLearn, Elements of AI (French version)
- Google for Developers, Machine Learning Crash Course: prerequisites and prework
- fast.ai, Practical Deep Learning for Coders
- Hugging Face, LLM Course
- Kirkpatrick Partners, The Kirkpatrick Model
Frequently asked questions
Do engineers need to learn Python to work with AI?
Most do not. Operators, managers and most engineers need AI literacy: what a model or assistant can do, how to check its output and which data may leave the site. Engineers who will prepare data and build models do need Python and basic statistics, and the free courses listed in this guide state those prerequisites.
How long does it take to train an engineer in AI?
It depends on the role. Literacy can be taught in short sessions before a tool goes live, on the tool itself. Becoming a practitioner who builds and maintains models takes months of courses and project work on the plant's own data, with time protected from operational duties.
Should we train our own engineers or hire AI graduates?
Usually both. Graduates, such as those of Algeria's National Higher School of Artificial Intelligence, bring the methods, while your engineers bring knowledge of the process, the tags and the equipment. Pairing them on a real project transfers skills in both directions.
Can our engineers use public chatbots such as ChatGPT at work?
Only within written rules. Decide which tools are approved and which documents and data must never be typed into a public service, and teach these rules in the literacy module. For confidential procedures, reports and process data, use an assistant hosted on your own infrastructure.
What training should we require from an AI vendor?
Name the people to be trained and the hours of hands-on training, require documentation in the languages your teams use, and include a shadowing period. Make acceptance depend on your team retraining or redeploying a model without the vendor. For Algerian public tenders, Article 60 of Law 23-12 already requires the specifications to provide, as applicable, for training and the transfer of know-how.
How do we stop operators from losing their manual skills?
Keep the procedure for operating without the AI in the SOPs and practise it regularly. The joint guidance on AI in operational technology led by CISA warns that heavy reliance on AI can erode the manual skills needed during AI failures, and recommends training teams to validate AI outputs, for example against other sensors.



