Industries16 min read

AI Fluency Training for Engineering and Product Teams

Dilr Academy is the AI tutor from DILR.AI that builds courses and teaches to mastery, and this guide applies that lens to AI fluency training for engineering, product and customer success teams: what each role should learn, how to measure fluency by behaviour, and how to run the first quarter.

On this page 9 sections
AI Fluency Training for Engineering and Product Teams DILR ACADEMY · SAAS AND TECHNOLOGY AI Fluency Training for Engineering and Product Teams 66 % of developers asked cite AI answers that are almost right Source: Stack Overflow Developer Survey 2025 dilr.ai/blog

Most UK SaaS companies already have AI inside the engineering workflow. Coding assistants sit in the editor, a support copilot drafts replies, and product managers summarise research with a chat tool. The tools arrived fast. The skill to judge what they produce did not, and that gap is now the expensive part. The 2025 DORA report from Google Cloud found that 90% of nearly 5,000 technology professionals surveyed use AI at work, while 30% report little or no trust in the code it generates.

This guide is written for the Head of Learning and Development at a UK software or technology company, and for the COO who signs off the budget. It sets out what AI fluency means for engineering, product and customer success roles, how to measure it by what people do rather than by completions, and how to run the first quarter of a programme. It deliberately stays narrow. The legal duty under the EU AI Act sits in our Article 4 AI literacy guide, the sector view of where AI pays sits in our UK SaaS and technology hub, and the platform buying criteria sit in our AI teacher buyer's guide. This post covers the curriculum and the measurement in between.

This guide is shipped by the team behind Dilr Academy, an AI tutor that builds interactive, multilingual courses and moves on only once the learner has understood. Or see DATS, our five-stage system for placing AI where it changes a P&L.

What does AI fluency mean for an engineering or product team?

AI fluency for an engineering or product team is the practised ability to use AI tools inside real work, judge whether an output is correct, and decide when a human must take over. It differs from awareness training, which explains what AI is. A fluent engineer can review a generated change as critically as a colleague's, and a fluent product manager can tell a sourced summary from a confident guess.

The distinction matters because most technology teams already pass the awareness bar. In the Stack Overflow Developer Survey 2025, 84% of respondents said they use or plan to use AI tools in their development process, up from 76% a year earlier, and 51% of professional developers use them daily. Usage is no longer the problem a programme has to solve. Judgement is.

A useful working definition has three parts. First, operation: the person can run the tools the company has approved, configure them for the task and work within the context limits. Second, evaluation: the person can check an output against a source, a test or a specification, and can explain why it is wrong when it is. Third, escalation: the person knows which decisions the tool may never close on its own and routes them to the right human. Our AI operating model work treats that third part as a decision-rights question, which is why fluency and governance are designed together rather than in separate projects.

None of this replaces deep machine learning skills for the small group who build models. Fluency is the layer every engineer, product manager and customer success lead needs once AI sits in the daily workflow, in the same way that every engineer needs to read a stack trace whether or not they wrote the runtime.

Why do engineering teams need AI fluency training after a tool is deployed?

Engineering teams need AI fluency training after deployment because the risk moves downstream once AI output starts flowing. Before a rollout the question is access. After it, the question is whether reviewers, testers and product owners can absorb a larger volume of generated work without lowering their standards. A SaaS company that trains only at launch teaches people to use the tool, then leaves them alone with its output.

The evidence on that downstream shift is consistent. Faros AI's Speed Trap report, its Q3 2026 analysis of telemetry from 22,000 developers across 4,000 teams, found pull requests merged without any review up 76.3%, time in QA up 300.6% and monthly incidents up 125.4%, even though incidents per pull request rose only 14.5%. Faros cautions that findings such as its agentic-review result are correlational, so read the figures as patterns across those teams rather than as the effect of a single tool. The pattern is still the one a training lead should design for: authoring got faster, and the bottleneck moved to the people who check.

The trust data points the same way. In the same Stack Overflow survey, 46% of developers said they actively distrust the accuracy of AI tools against 33% who trust it, and only 3% said they highly trust the output. Distrust without a method is not fluency either. A reviewer who distrusts everything slows the team down; a reviewer who trusts everything ships defects. The skill a programme builds is calibrated trust: knowing which outputs to check closely, which to sample, and which to reject.

DORA's own summary of the 2025 research puts the organisational point plainly. As Nathen Harvey and Derek DeBellis wrote when announcing the 2025 DORA report, "AI doesn't fix a team; it amplifies what's already there." DORA here is the DevOps Research and Assessment programme run with Google Cloud, not the EU financial regulation that shares the initials. For a training lead, the amplifier finding means that fluency work is not optional polish on an AI rollout: it decides whether the rollout amplifies good review habits or weak ones.

What developers find frustrating about AI tools
66%Almost right45.2%Slower debugging20%Less self-confidence16.3%Hard to follow
Share of respondents to the Stack Overflow 2025 frustrations question citing each problem; more than one answer allowed. Source: Stack Overflow Developer Survey 2025, AI section

Read the chart as a curriculum brief. The largest frustration, cited by 66% of respondents to that question, is output that is almost right but not quite. That is an evaluation problem, and it is teachable. The third bar matters too: 20% said they had become less confident in their own problem-solving, which is the case for building fluency that keeps the human skill alive rather than outsourcing it.

What should each role learn: engineering, product and customer success?

Each role in a SaaS company should learn the AI skills that match the decisions it owns. Engineers need to review, test and constrain generated code. Product managers need to judge AI-produced research and specify AI features precisely. Customer success needs to check AI-drafted answers against the product and the contract. One generic course for all three teaches the tool and misses the judgement each role is accountable for.

The table below is our working split. It is a starting point to adapt, not a standard, and it assumes the company has already chosen its approved tools, often through work such as our AI solutions practice.

RoleOperateEvaluateEscalate
EngineeringUse the approved assistant inside the repository, with scoped context and the team's conventionsReview generated changes against tests and specifications; spot plausible but wrong logic; size changes for reviewSecurity-sensitive code, data migrations, anything touching customer data or billing
ProductSummarise research and support tickets; draft specifications with AI helpCheck every claim in a summary against its source; separate observed user behaviour from model inferenceAI feature decisions that change what a customer is told, charged or shown
Customer successDraft replies and account summaries with the support copilotCheck drafted answers against product documentation and the customer's contractCommitments on price, refunds, data handling or roadmap dates

Two habits cross all three columns. The first is source discipline: every AI-assisted claim that leaves the team carries the source it came from, or it does not leave. It is the habit we would teach first, and the team at Dilr.ai can explain why. The second is size discipline: generated work is broken into pieces small enough for a human to review properly. Faros reported average pull request size up 71.8% alongside the rise in unreviewed merges, which is why size belongs in the engineering curriculum rather than in a style guide.

The same diagnostic logic underpins our AI placement diagnostic, which maps where AI belongs in a workflow before anyone is trained to use it there. Training a team on a use case that should not exist is the most expensive form of fluency work.

Engineering leads who want a deeper treatment of the review side can read our post on review queues for AI agent work, and our engineering note on evaluation harnesses covers the test infrastructure that makes evaluation skills useful at scale.

How do you measure AI fluency without counting course completions?

You measure AI fluency by sampling the work people produce with AI and scoring the judgement in it, not by counting completed modules. A completion record proves attendance. A sample of reviewed pull requests, checked summaries and drafted replies proves whether people catch errors, cite sources and escalate correctly. Learning and development teams that report completions alone cannot tell a fluent team from a busy one.

A practical scorecard uses four signals, each drawn from work the team already does.

  1. Defect catch rate in review. Seed a small number of known faults into AI-generated changes in a practice repository and measure how many reviewers find. Repeat each quarter.
  2. Source discipline in written work. Sample product summaries and customer replies and check what share carry a verifiable source for each factual claim.
  3. Escalation accuracy. Run short scenarios in which the right answer is to stop and route the decision to a named owner, and score whether people do. Clear decision rights for AI use make this signal meaningful.
  4. Self-reported confidence against measured skill. Ask people how confident they are, then compare with the first three signals. A large gap in either direction is a coaching target.

The fourth signal exists because self-reports and outcomes can diverge. Two separate sources illustrate the risk: in the DORA 2025 survey more than 80% of respondents believed AI had increased their productivity, while Faros's telemetry, a different dataset, showed time in QA up 300.6%. Neither proves the other wrong, but together they are a reason not to take sentiment as a programme's outcome measure.

The teaching method matters as much as the measure. A course that asks the learner to reason before it explains builds the evaluation habit the scorecard tests. Our explainer on the Socratic AI tutor sets out how that question-first loop works, and our guide to measuring AI tutor engagement in a pilot covers which learner signals are worth tracking and which are noise.

Where does the Skills England AI skills framework fit for a technology company?

The AI skills framework in the report co-developed with Skills England gives a technology company a shared vocabulary rather than a ready-made curriculum. It groups AI skills into technical, responsible and ethical, and non-technical domains. For a SaaS training lead, its most useful contribution is the sector finding that digital and technology teams have gaps in using low-code tools, explaining AI outputs, designing for inclusion and using AI responsibly in products.

The report, AI skills for the UK workforce, was published on 29 October 2025 and updated on 4 November 2025. It states that its opinions are those of its author rather than official Skills England policy, and that its workshop findings are qualitative and not designed to be statistically representative. Read it as structured expert input, not as a benchmark.

For the Digital and Technology sector, the report lists the skills gap areas as using low-code tools, explaining AI outputs, designing for inclusion and using AI responsibly in products. It lists the main barriers as training that is too technical, poor support for women and non-technical staff, and limited options for older workers, career returners and people outside the main hubs. That second list is a direct challenge to the default SaaS approach of handing everyone a vendor course written for engineers. The tools package makes the same point in its own words: "Stakeholders stressed that technical AI skills should be taught in context."

We use the framework's three domains as a cross-check on the role table above rather than as its replacement. Operate maps mostly to the technical domain, evaluate and escalate map mostly to the responsible and ethical domain, and the habit of explaining an output to a colleague or a customer sits in the non-technical domain. A programme that covers all three columns for each role is, in practice, covering all three domains. How a training provider maps a full course catalogue to the framework is a separate question, and one we leave to a dedicated piece.

The national context gives the framework weight. The government's AI Skills Boost programme now aims to upskill 10 million UK workers in AI skills by 2030, increased from the original 7.5 million target announced in June 2025. The Department for Science, Innovation and Technology estimates that the UK digital sector's gross value added grew 152.2% in real terms, from £69 billion in 2010 to £174 billion in 2024. A sector of that size will set the pace for AI skills in the wider economy whether or not its training is any good.

How should a SaaS company run an AI fluency programme in its first quarter?

A SaaS company should run the first quarter of an AI fluency programme as a measured pilot inside one team, not as a company-wide launch. Baseline the scorecard, teach the role-specific evaluation skills against the team's own codebase and tickets, re-measure, and only then widen. A quarter is long enough to see behaviour change in review and short enough to stop a programme that is not working.

A first-quarter AI fluency programme
01BaselineWeeks 1 to 202Teach in contextWeeks 3 to 803Re-measureWeeks 9 to 1004Decide and widenWeeks 11 to 13
Each stage ends with a measurement before the next begins.

Weeks 1 to 2, baseline. Pick one team that already uses AI daily; if the choice is unclear, an AI placement diagnostic shows where AI already sits. Usually that is a product squad with its engineers, product manager and the customer success lead who supports its customers. Run the four scorecard signals before any teaching starts. Record the results privately: the baseline is for the programme, not for performance reviews, and people who fear it will be used against them will not show their real habits.

Weeks 3 to 8, teach in context. Teach evaluation against the team's own material: real pull requests with seeded faults, real tickets, real specifications. The tools package's line that technical AI skills should be taught in context is the design rule here. Short, frequent sessions inside the working week beat a two-day course, because the habit being built is a daily one.

Weeks 9 to 10, re-measure. Run the same four signals again. Look for movement in defect catch rate and source discipline first; escalation accuracy often lags because it depends on decision rights being clear, which is an operating model question as much as a training one and part of our DATS methodology.

Weeks 11 to 13, decide and widen. If the signals moved, extend to the next two teams and assign the first team's strongest reviewers as peer coaches. If they did not, change the teaching before adding people. Our AI execution office exists for companies that want this cadence run with them over several quarters rather than handed over as a plan.

What is the best AI fluency training for engineering teams in 2026?

The best AI fluency training for engineering teams in 2026 is the option that teaches evaluation against your own work and measures behaviour afterwards, judged on four criteria: role specificity, teaching method, measurement and language reach. No single provider wins on every criterion. Large course libraries win on breadth, programme providers win on structure, and a tutor that builds courses from your curriculum wins on context.

Applied to the main options, the criteria give a scoped verdict rather than a ranking.

Course libraries. Udemy Business advertises more than 30,000 courses in AI, technology and other subjects, and Coursera for Business describes its offer as learning solutions and certification programmes for teams. For a company that wants broad, self-paced access across many roles at once, a library is the sensible choice and it will beat a custom build on breadth every time. Its weakness for fluency work is context: a generic course cannot use your repository or your tickets, the context gap our UK SaaS sector guide also flags.

Adoption and programme providers. Multiverse now positions itself as an AI adoption layer, and Section AI as an enterprise AI training and adoption platform. Providers in this group position themselves around adoption across a whole organisation rather than around single courses. Where a company wants an outside provider to run organisation-wide adoption for it, one of these is likely the better fit than a tutor.

A tutor that builds from your curriculum. Dilr Academy builds interactive, multilingual courses from the curriculum of the organisation that licenses it, with live diagrams, simulations and Socratic questioning, and it tracks mastery with knowledge tracing so it moves on only once the learner has understood. Its built courses already include Python, at 40 taught hours, and a generative AI course of 8 modules and 41 topics. It is licensed per institution to universities, colleges and training providers today and is not sold to individual learners. A technology company that owns its own fluency curriculum, or a training provider that serves technology companies, should talk to us about fit before assuming one way or the other.

Our introduction to Dilr Academy explains how the tutor teaches, and the comparison linked at the end of this guide sets the wider field side by side.

Does the EU AI Act require AI fluency training for engineers?

The EU AI Act does not require a specific AI fluency course for engineers. Its Article 4 AI literacy duty binds providers and deployers of AI systems, and after the 2026 Digital Omnibus amendment it asks them to take measures to support the development of AI literacy rather than guarantee any level. A UK SaaS company comes within the Act's reach where it places an AI system on the EU market or the system's output is used in the EU.

Our guide to Article 4 AI literacy covers who the duty binds, how the amendment changed it, and how to evidence proportionate compliance. The practical overlap with this guide is evidence: a fluency programme that records what was taught, to whom and with what measured result is also the record an AI literacy review would ask to see. Other EU AI Act obligations, such as the transparency duties for chatbot features, are separate product obligations and are not satisfied by training alone.

Can Dilr Academy teach a SaaS company's own AI curriculum?

Dilr Academy is licensed per institution to the organisation that owns the curriculum, and today that means universities, colleges and training providers. It builds interactive, multilingual courses from that curriculum and teaches each learner one to one with Socratic questioning and mastery tracking. A SaaS company with its own AI fluency curriculum should discuss fit on a call rather than assume a licence model applies.

The courses Dilr Academy builds use validated, templated interactive widgets rather than generated layouts, run interactive elements in a sandbox, and keep each institution's learners and progress isolated from every other institution's. Those properties matter to a technology buyer whose security review will ask how learner data is separated. A scoping call is the quickest way to test fit, and the AI teacher buyer's guide linked in the introduction lists the questions to ask any AI tutor before a pilot.

Want to see how this would work in your organisation? Explore Dilr Academy, compare the best AI education tools of 2026, see our DATS methodology, or read about our approach to placing AI inside enterprise systems. More sector guides sit in our industries collection.

Talk to the operators

Teach the judgement, not just the tool.

30-min scoping call · No deck · Confidential. We will tell you where AI fluency work will move your review and support quality, and where it will not.

Written by the Dilr.ai engineering team, practitioners who ship enterprise AI in production. Follow us on LinkedIn for shipping notes, or subscribe via the RSS feed.

  • ai fluency training engineering teams
  • saas technology
  • ai upskilling saas company
  • ai training product managers uk
  • corporate ai literacy programme tech
  • ai training for developers reddit
  • best ai training for engineering teams 2026
  • dilr academy

What does this article answer?

The questions this article answers, in short

  • AI fluency for an engineering or product team is the practised ability to use AI tools inside real work, judge whether an output is correct, and decide when a human must take over. It differs from awareness training, which explains what AI is. A fluent engineer can review a generated change as critically as a colleague's, and a fluent product manager can tell a sourced summary from a confident guess.
  • Engineering teams need AI fluency training after deployment because the risk moves downstream once AI output starts flowing. Before a rollout the question is access. After it, the question is whether reviewers, testers and product owners can absorb a larger volume of generated work without lowering their standards. A SaaS company that trains only at launch teaches people to use the tool, then leaves them alone with its output.
  • Each role in a SaaS company should learn the AI skills that match the decisions it owns. Engineers need to review, test and constrain generated code. Product managers need to judge AI-produced research and specify AI features precisely. Customer success needs to check AI-drafted answers against the product and the contract. One generic course for all three teaches the tool and misses the judgement each role is accountable for.
  • You measure AI fluency by sampling the work people produce with AI and scoring the judgement in it, not by counting completed modules. A completion record proves attendance. A sample of reviewed pull requests, checked summaries and drafted replies proves whether people catch errors, cite sources and escalate correctly. Learning and development teams that report completions alone cannot tell a fluent team from a busy one.
  • The AI skills framework in the report co-developed with Skills England gives a technology company a shared vocabulary rather than a ready-made curriculum. It groups AI skills into technical, responsible and ethical, and non-technical domains. For a SaaS training lead, its most useful contribution is the sector finding that digital and technology teams have gaps in using low-code tools, explaining AI outputs, designing for inclusion and using AI responsibly in products.
  • A SaaS company should run the first quarter of an AI fluency programme as a measured pilot inside one team, not as a company-wide launch. Baseline the scorecard, teach the role-specific evaluation skills against the team's own codebase and tickets, re-measure, and only then widen. A quarter is long enough to see behaviour change in review and short enough to stop a programme that is not working.
  • The best AI fluency training for engineering teams in 2026 is the option that teaches evaluation against your own work and measures behaviour afterwards, judged on four criteria: role specificity, teaching method, measurement and language reach. No single provider wins on every criterion. Large course libraries win on breadth, programme providers win on structure, and a tutor that builds courses from your curriculum wins on context.
  • The EU AI Act does not require a specific AI fluency course for engineers. Its Article 4 AI literacy duty binds providers and deployers of AI systems, and after the 2026 Digital Omnibus amendment it asks them to take measures to support the development of AI literacy rather than guarantee any level. A UK SaaS company comes within the Act's reach where it places an AI system on the EU market or the system's output is used in the EU.
Written by Dilr.ai EngineeringEngineering team

Can voice AI handle the calls in my sector?

Dilr Voice answers and places calls for regulated sectors

From clinics and estate agents to financial services, Dilr Voice handles calls 24/7 with the consent, recording and audit rules each sector needs.

What should I read next?

Every article in Industries