Human-Centred AI Learning
Insights from 14 conversations about work and organizational change
People doing the work should help shape the change
AI can help people prepare a document, organize information or develop an idea. Making those benefits useful across an organization brings further questions: how work is coordinated, how people learn, who checks quality, and who decides what happens next.
This report examines those questions through 14 exploratory conversations with leaders and specialists. It connects their experiences with selected research and offers a practical approach to learning about AI together.
Our position is that people using AI, and people affected by its use, should have a meaningful role in understanding and shaping changes to their work.
The report in brief: AI learning includes learning how work changes
The conversations describe useful applications, uneven experiences and choices that remain unresolved. Some participants are already building thoughtful support. Others are working through gaps between an available tool and a workable way of using it.
A faster task is only part of the work
Participants reported substantial savings on specific activities, including meeting minutes and early drafts. Those accounts do not establish the net benefit across a workflow. Preparation, checking, handoffs and coordination still need attention. Someone must be able to act on what a team learns.
Finding 1 →02People need a route into decisions
Feedback arrangements exist in several accounts, but being asked for input does not establish what that input can change. Participants disagreed about how much authority staff should have. Clear decision responsibilities and a response to concerns are central to a credible invitation to participate.
Finding 2 →03Learning needs relevant practice and judgment
Training resources alone do not provide time to learn or show how a tool fits a particular job. Participants described useful demonstrations, peer exchange and trials with different groups. They also emphasized the expertise needed to assess AI output.
Finding 3 →04Governance happens in everyday situations
An approved tool does not settle every question about consent, information sharing or responsibility. People need workable guidance and somewhere to take situations the rules do not resolve.
Finding 4 →05Organizations choose how to use the gains
Participants described different possibilities: more output, better service, innovation, changes to staffing and shorter working time. Some were reported decisions; others were aspirations. The effects on work depend partly on those choices.
Finding 5 →Include the people doing the work in examining these questions. Give them time to compare experiences, make sense of differences and test changes with the colleagues and specialists involved. Connect that learning to decisions about the work itself.
Does your approach to AI learning give people a way to examine and influence how their work is changing?
Begin with people and their experience
HumLab is a human systems consulting firm. We wanted to understand how people were making sense of AI in their organizations: what was helping, what was difficult, and what support they thought would be useful. Fourteen conversations gave us accounts of practical benefits, active experimentation, disagreement and uncertainty.
To the people who took part: your contributions gave this inquiry its substance. You described practices worth learning from, corrected assumptions in our questions, and challenged ideas about what organizations need. Those differences belong in the report.
“A lot of times we're engaged in these discussions with consultants who already have a stance — it's not a curiosity.”
That challenge matters for how we work. An inquiry loses its value if the answer is decided before people speak. Some participants wanted specialist help with particular applications. Others questioned whether AI was useful for the problem at hand. These are different needs.
What human-centred change means to HumLab
People inside an organization have expertise in how their work actually happens. Their knowledge includes informal coordination, exceptions, professional standards and consequences that may be difficult to see from outside their role. Bringing those accounts together can reveal dependencies that no one person sees in full.
Our role is to help people investigate a shared problem, examine assumptions and turn what they learn into decisions and practical tests. Leadership direction, technical knowledge and lived experience each have a place. Participation needs clarity about what people can influence and who is accountable for the final decision.
Who took part
Fifteen people took part in total. Fourteen completed the survey and fourteen were interviewed, with thirteen participating in both. One survey respondent was not interviewed because the interview could not be rescheduled, and one interviewee did not complete the survey. All fourteen completed survey responses are included in the survey results.
Survey respondents
Respondents were senior people close to AI adoption: HR, learning and organizational development leaders; technology and AI leads; chief executives; and practising professionals. Organizations were split between small (fewer than 50 people, five organizations) and large (2,000 or more, four organizations), with five in between. Nonprofits and professional services were the most represented sectors.
Organization size
Q1. How many people work at your organization?
| Fewer than 50 | 5 |
|---|---|
| 50–199 | 1 |
| 200–499 | 2 |
| 500–1,999 | 2 |
| 2,000 or more | 4 |
Sector
Q2. What sector best describes your organization?
| Not-for-profit or foundation | 4 |
|---|---|
| Professional services | 3 |
| Healthcare | 2 |
| Technology / software | 2 |
| Financial services | 1 |
| Retail or consumer | 1 |
| Education | 1 |
Respondent role
Q3. Open-text titles, grouped by HumLab
| People, HR, L&D or OD | 5 |
|---|---|
| Technology, AI or digital | 4 |
| Executive leadership | 3 |
| Legal or consulting practice | 2 |
Interview participants
Fourteen semi-structured interviews were held by video between May 6 and September 24, 2026, ten in May, three in July and one in September. Two HumLab researchers conducted them. Most ran 30 to 60 minutes. To protect anonymity, each participant has a random identifier (I01 to I14) and is described only by a broad role family and sector. Quotations are anonymized; their source identifiers are retained in an internal evidence register rather than displayed publicly.
| ID | Role family | Sector |
|---|---|---|
| I01 | People & learning | Health and benefits |
| I02 | Technology & AI | Software |
| I03 | People & learning | Higher education |
| I04 | Executive leadership | Software services |
| I05 | People & learning | Financial services |
| I06 | Technology & AI | Nonprofit |
| I07 | Professional practice | Legal services |
| I08 | Technology & AI | Retail |
| I09 | Executive leadership | Nonprofit |
| I10 | People & learning | Health technology |
| I11 | Professional practice | Management consulting |
| I12 | Executive leadership | Nonprofit sector support |
| I13 | Professional practice | HR and payroll services |
| I14 | People & learning | Technology |
Interviewees included five people and learning leaders, three technology and AI leaders, three executives and three practising professionals. Their organizations were at very different stages: one had hired its first AI lead a month earlier, while another was “pretty AI forward ... more forward than I've ever seen.”
What the survey shows
The charts below show the number of respondents choosing each answer (n = 14 throughout). Grey bars show “not sure”, “too early to tell” or no answer. Answer options that no one chose are not shown. With 14 respondents, each answer represents one organization's perspective as reported by one senior person, so small differences should not be over-read.
Where organizations are
Most respondents report that they are past the starting line. Nine of 14 have been using AI for a year or more, and seven describe use as widespread or reaching most staff. Six describe their approach as a coordinated mix of tool-led, leadership-led and bottom-up adoption. One respondent said their organization had “not started yet”, yet also reported that most staff already use AI at least occasionally. This pattern of unofficial use running ahead of official adoption also emerged in the interviews.
Time using AI
Q4. How long has your organization been actively using AI tools day to day?
| Not started yet | 1 |
|---|---|
| Less than 6 months | 2 |
| 6 months to 1 year | 2 |
| 1 to 2 years | 6 |
| More than 2 years | 3 |
Breadth of use
Q5. How widely are AI tools being used by staff?
| Almost no one yet | 1 |
|---|---|
| A small group of early adopters | 2 |
| Some teams use them regularly | 3 |
| Most staff use them occasionally | 2 |
| Widespread and embedded | 5 |
| Not sure | 1 |
Approach to adoption
Q6. How would you describe the way your organization is approaching AI adoption?
| Mostly tool-led | 3 |
|---|---|
| Mostly leadership-led | 2 |
| Mostly bottom-up | 1 |
| A coordinated mix of all three | 6 |
| Not sure | 1 |
| No answer | 1 |
Listening and voice
Only five organizations gather information about staff experience through structured or systematic channels, and three have not yet gathered any. Only one says frontline staff are regularly involved in shaping AI decisions. Four respondents were not sure whether frontline staff have a voice at all.
Half the respondents (7 of 14) see a moderate or significant gap between how leadership talks about AI and how people doing the work experience it. Only one sees no real gap.
Listening to staff experience
Q7. Has your organization gathered information about how staff experience AI?
| Not yet | 3 |
|---|---|
| Informally, through conversations | 4 |
| Some structured channels | 3 |
| Systematically, as part of rollout | 2 |
| Not sure | 2 |
Frontline voice
Q8. Are frontline staff given a voice in how AI is being adopted?
| Not asked for input | 1 |
|---|---|
| Informal opportunities to give feedback | 4 |
| Consulted occasionally | 4 |
| Regularly involved in shaping decisions | 1 |
| Not sure | 4 |
Leadership–experience gap
Q10. Gap between how leadership talks about AI and how people experience it
| No real gap | 1 |
|---|---|
| A small gap | 3 |
| A moderate gap | 3 |
| A significant gap | 4 |
| Not sure | 3 |
Building capability
Nine of 14 respondents reported that their organizations have invested minimally (mostly informal pointers and tips) or not at all in AI skills. Five report substantial or comprehensive investment. Views on managers are evenly split: five say managers are mostly or very well-equipped, four somewhat equipped and five not very or not at all equipped.
All three respondents who reported only a small leadership–experience gap also reported substantial investment in skills. All four who reported a significant gap reported minimal or no investment. In this small sample, greater investment in AI skills was associated with a smaller reported gap between leadership messaging and staff experience. The data cannot establish causation or whether this pattern would hold across a larger sample.
Investment in AI skills
Q9. Has your organization invested in helping staff use AI well?
| Not at all yet | 2 |
|---|---|
| Minimally: informal pointers and tips | 7 |
| Substantial investment | 3 |
| Comprehensive, ongoing | 2 |
Managers’ readiness
Q12. How well-equipped are managers to lead their teams through AI adoption?
| Not at all equipped | 1 |
|---|---|
| Not very well-equipped | 4 |
| Somewhat equipped | 4 |
| Mostly well-equipped | 4 |
| Very well-equipped | 1 |
How it is going
Half rate adoption as mostly successful, and five say it is too early to tell. Five are confident that the time and money invested in AI are producing the expected value, five are neutral, and two are skeptical. Three of the seven who rated adoption mostly successful also reported a moderate or significant gap between leadership talk and staff experience. Success at the organizational level and the experience on the ground are not the same thing.
Reported friction is low to moderate in most organizations. Four hear none and four hear a little. Two hear a lot.
Perceived success
Q11. Overall, how successful has AI adoption been so far?
| Mixed results | 2 |
|---|---|
| Mostly successful | 7 |
| Too early to tell | 5 |
Confidence in value
Q13. Is the time and money invested in AI producing the value it should?
| Very skeptical | 1 |
|---|---|
| Somewhat skeptical | 1 |
| Neutral | 5 |
| Somewhat confident | 3 |
| Very confident | 2 |
| Not sure | 2 |
Staff friction
Q14. How much friction or pushback are you hearing from staff about AI?
| None | 4 |
|---|---|
| A little | 4 |
| A moderate amount | 2 |
| A lot | 2 |
| Not sure | 2 |
The most common staff reaction, in respondents' words
Nine respondents answered the open-text question about the most common staff reaction they hear. Five described practical value or appetite: help with emails and report summaries, feeling “better equipped and supported”, curiosity and excitement, and a wish for “more and better tools”. One respondent added: “it saves me time, but I'm only using a small percentage of its capabilities.” One said the most common question is simply “How do I use it more effectively?”
Four described concern. One named the elimination or change of job roles. One named dislike of change and the extra time it takes to learn something new. One named “questions on ethics and lack of choice.” One described a wide spectrum running from mistrust and skepticism (about data collection, data security and consent) through fear (of layoffs, environmental impact and unregulated development) to outright rejection, rooted in the politics of the companies building AI and in questions of Canadian digital sovereignty. These reactions are explored in the interviews below.
Five findings from the conversations
Improving work requires more than faster tools
One participant reported that preparing board minutes had fallen from three or four hours to between 30 and 60 minutes. The tool also helped them be more present in the meeting. Human checking remained part of the task. This was a clear, bounded benefit in their account, rather than a measured return across the organization.
Another participant described generating an initial learning storyboard in minutes where a draft had previously taken days. Editing still followed. Faster draft production did not mean that a finished learning product required only minutes.
The surrounding process can become the problem
A participant described administrative work spread across disconnected platforms owned by different departments. The desired improvement depended on those departments connecting their systems and agreeing how the work would run. It might require ordinary systems integration; AI was not necessarily the missing component.
In a separate account, an early phase of a processing-automation rollout increased complexity and backlog. The participant said changes to processes and job aids had been underestimated. Trainers and change support were already involved. The difficulty was therefore more specific than an absence of training or attention to change.
“When you change the tool, it changes the processes.”
That implementation involved optical character recognition and process automation within a broader AI discussion. It is one account, without independent service data, and cannot establish what happens in most generative-AI rollouts. It does show why the surrounding work deserves examination.
Look for the person who can connect the work
Other accounts described constructive coordination: product ownership linking business priorities to technical work, and a people-and-culture specialist contributing to a technical rollout through useful projects. Organizations may already have this capacity. The practical issue is whether responsibility, resources and access to the right expertise come together around the problem.
Examine the complete workflow
“Is AI the best way? If so, how do we use it?”
A participant wanted support from people who understood the function and had implemented relevant applications. We take that request seriously: facilitation can help identify the problem, while technical and domain expertise are needed to judge particular solutions.
A practical starting point
HumLab proposes bringing together people who produce, check and receive the work. Map one recent example and examine where effort moves. The diagram below is a discussion aid, not a result from the interviews.
- PrepareWhat information, permissions and context are needed?
- ProduceWhat does AI do, and what does a person still contribute?
- CheckWho can judge quality, correct errors and stop unsuitable work?
- Hand overWhat does the next person need, and what extra work reaches them?
- Use and reviewDid the result improve the service or decision? Who can change the process?
Choose an outcome that matters to the work: usable quality, service reliability, total effort or time to completion. Review it with the people affected, alongside any new burdens. Tool availability and usage counts answer different questions.
People doing the work need a say
Several participants described ways their organizations seek employee input, including surveys, office hours, conversations with managers and shared channels. The challenge is not necessarily a lack of opportunities to speak up. It is whether employees understand how their feedback is considered, what decisions they can influence, and what happens after they raise a concern or suggest an improvement.
One participant initially described limited formal feedback, then recalled a recurring route to the chief executive as well as conversations with managers. Another described several listening mechanisms during a wider organizational change, while practical consequences for roles and daily work remained unresolved. Having channels and knowing what input can influence are different things.
There is a real disagreement about authority
A participant questioned whether staff should have the same degree of input into technical-risk choices that they expected in other areas. The account included many conversations with staff and a preference for clear leadership direction. It challenges the idea that the answer is simply more consultation.
“I'm not so sure that staff needs to have the level of input that they've been used to having in other areas.”
Another participant wanted to understand people's needs and explain tool decisions even when an individual preference was not adopted. These views point toward a useful distinction: a person can contribute knowledge and challenge a proposal without holding final authority over every choice.
Caution is information to investigate
One participant found writing assistance useful but wanted workflow automation that the available tool did not provide. They did not want to become a workflow developer. More encouragement to experiment would not, by itself, supply the missing capability.
Another combined selective AI use with objections about values and potential harms. A third corrected an interviewer's description of their reaction as anxiety. These accounts caution against deciding in advance what reluctance means. Lack of time, limited usefulness, professional responsibility and disagreement about purpose call for different responses.
Make influence specific
HumLab's position is that people affected by a change should help define the problem and examine proposed solutions. An invitation to participate should explain which choices are open, what constraints apply, and how decisions will be communicated.
Agree on the decision process before asking for input
| Question | A useful agreement |
|---|---|
| What problem are we solving? | People doing and receiving the work help describe it, using actual examples. |
| What can change? | Name the choices still open, such as workflow, support, tool use or the pace of implementation. |
| What boundaries apply? | Relevant specialists explain requirements and the reasons for them. Identify who can clarify or reconsider a constraint. |
| Who decides? | Name the accountable person or group and how employee experience will inform the decision. |
| What happens afterward? | Share the decision, its reasons and how concerns or unintended effects will be revisited. |
HumLab discussion aid. Decision responsibilities must be adapted to the organization and the issue.
Listen for differences without requiring agreement
In a facilitated conversation, ask people to describe a specific incident before assigning a label to it. Compare the accounts, including the experience of someone who receives the output or handles exceptions. Invite people to correct the interpretation. Where power differences make open discussion difficult, provide another way to contribute and be clear about confidentiality.
The useful outcome is an informed decision with visible follow-through. Agreement on every point is not a condition for learning together.
Make learning part of the work
Participants described training, expert sessions, approved tools, champions and informal demonstrations. The gap was often between access to support and an opportunity to use it on a relevant problem.
“So they made the resources available, but they don't necessarily allow you the time to study.”
This participant described learning resources competing with delivery commitments. Their account spanned professional settings; it should not be read as a statement about one employer. It nevertheless makes a concrete distinction: a resource can be available while time to use it remains scarce.
Useful learning was often close to an actual task
One participant described sitting with colleagues to demonstrate a shared tool. Another described short recordings of a technical leader working through problems and experienced users sharing their work. A third tested guidance with different pilot groups before wider rollout.
“I'm a big believer in doing it [...] you have to put it into reality.”
These are accounts of practice, not controlled evaluations of which method works best. They suggest practical questions: can people try the guidance, see where they become stuck and help improve it? Does support meet different levels of experience?
One participant found office hours useful for beginners but less helpful when the same introductory questions dominated. Another wanted learning tied to departmental problems after initial general training. A single session or resource cannot be assumed to meet all those needs.
HumLab recommends choosing a real task, providing time to work on it and reviewing the experience with colleagues. If the obstacle is a missing feature, an unclear rule or a decision outside the team's authority, record it as such and route it to someone who can act.
Build the ability to judge the output
Faster production leaves people with decisions about whether an output is accurate, appropriate and good enough to use. The board-minutes example still required checking. Another participant described inaccurate material in fundraising outputs. A consultant described using professional experience to frame a question and recognize an implausible answer while mentoring an intern.
Those accounts give human review a practical meaning. Someone needs to know what quality looks like, have access to the information needed to check it, and be able to act when something is wrong. Our recommendation is to include those conditions in learning design.
Practice making the judgment together
Use an appropriate, non-sensitive example. Ask a less experienced and a more experienced colleague to explain what they would accept, check or change in an AI output, and why. Compare their reasoning. Keep room for uncertainty and look for the knowledge that is missing.
Then decide what to try: better guidance, access to an expert, a revised checking routine or a different use of the tool. Review whether people can explain and improve their decisions on a later task. Speed and confidence alone will not answer that question.
Rules have to work in real situations
A participant described stopping AI note-taking in a session after attendees objected that they had not consented. Some content had already been captured. Another described discussions in which recording could support accessibility while conflicting with someone else's refusal to be recorded. Those discussions extended beyond that participant's own workplace.
The tension is practical. People can have legitimate, different needs in the same meeting. Knowing that a tool is approved does not answer whether its use is appropriate in that situation or what alternative arrangement is possible.
Useful experimentation needs support
One participant described a prototype that exposed sensitive technical information. The team detected and addressed the problem without blaming the individual, while recognizing the need for clearer safeguards. The same participant defended the value of experimentation. Their account combines enthusiasm with a specific lesson about the support required.
Other participants described governance already in place, including policy review, training and approved tools. Restrictions could also limit useful functions. These are questions about how guidance works in practice, alongside what the technology can safely do.
Work through a situation together
HumLab proposes using a real, appropriately de-identified dilemma with the people involved and the relevant specialists. For meeting notes, ask what information is captured, where it goes, who receives it and what choices participants have. Consider accessibility needs and an alternative way of achieving the meeting's purpose.
Agree who can pause the activity, who can clarify the rule and what happens to information already captured. If the situation exposes a gap in policy or support, assign someone to resolve it and communicate the result. Technical, privacy or legal questions should go to the people qualified to answer them.
This is learning through an organizational decision. The aim is guidance people can use, together with a clear route for situations that guidance does not yet cover.
Decide what happens to the time saved
Participants described different uses for greater capacity. One organization was piloting a shorter working week while seeking to maintain pay and output. Another wanted AI to create time for innovation. In a third account, a planned vacancy was not filled after some tasks were automated and the remaining work could be absorbed by the team.
These examples have different evidential status. The shorter-week pilot was an intention without evaluated results or evidence that AI alone enabled it. Time for innovation was a stated aim. The vacancy decision was reported as having happened; it was not a dismissal. Together, they show choices that deserve explicit discussion.
Expectations can change alongside the task
One participant described AI-assisted productivity becoming a managerial expectation. Others described pressure to move faster or uncertainty about the future value of specialist work. Positive accounts also included demanding work, intellectual enjoyment and opportunities to contribute in new ways.
The interviews do not establish that AI caused burnout, reduced overall employment or relieved workload. Existing commercial pressures, restructuring and management practices also shaped the accounts. Treating every outcome as an effect of AI would obscure the decisions people can actually examine.
Make the intended benefit explicit
Before raising output expectations, HumLab recommends examining the full effort required, including checking and work transferred to colleagues. Discuss how any capacity gained will be used and who will be affected. The decision might prioritize service quality, development time, new work or additional output; competing priorities should be named.
Ask what happens to workload, role clarity and opportunities to learn. Agree how the team will recognize an improvement and how it can raise a problem. A faster task does not, by itself, decide what a better working arrangement should be.
Learning from you and connecting learning to action
The 14 people who generously shared their experiences and perspectives with us have helped shape how we think about human-centred AI learning and integration. Their insights reinforced the importance of approaching AI not simply as a technological change, but as something that affects how people work, learn, make decisions and relate to one another.
Drawing on these conversations, alongside our own research and experience, we see value in approaches that bring people together to examine how AI fits their work and actively involve them in shaping what changes. This means combining structured inquiry, relevant research and attention to relationships and responsibilities. People inside an organization bring knowledge of their own system, while technical and domain specialists contribute where their expertise is needed.
The following principles reflect how this learning is informing our approach:
- 01
Start with a real problem
Choose a task, decision or difficulty people recognize. Establish what improvement would mean and who is affected. AI may be part of the answer; a process change or another tool may be more useful.
- 02
Bring different experiences together
Hear from people using the tool, those receiving its output, leaders and relevant specialists. Use concrete examples. Include experiences that challenge the preferred explanation, and provide ways to contribute when speaking openly is difficult.
- 03
Examine what produces the problem
Compare accounts and distinguish what happened from assumptions about why. Look at the tool alongside workload, information, incentives, relationships and authority. Return the interpretation to people for correction and discussion.
- 04
Choose a change people can test
Agree on a manageable experiment, its boundaries, the person responsible and what would count as improvement. Make sure the team has the time, support and authority to carry it out. Decide what would justify changing course or stopping.
- 05
Review and adjust
Compare the experience with the intended result. Look for extra work or unintended effects elsewhere. Keep, revise or stop the change, and share what others need to know. The next question should follow from what actually happened.
What the conversations suggest about learning together
Reflecting on the interviews and our subsequent discussions as a HumLab team, we have returned to the uncertainty that AI introduces into everyday ways of working. This is our interpretation of what we heard, rather than an additional interview finding: organizations may have very different levels of readiness, yet still need opportunities to ask what is changing, exchange practical experience and identify where decisions or safeguards are missing.
These conversations have also influenced how we think about making AI learning more social. A short inquiry into how people are experiencing AI can help surface different needs across roles or teams. Facilitated peer exchange can then allow colleagues to demonstrate useful practices, examine concerns and learn from one another, rather than leaving experimentation to isolated individuals. The right format and scale depend on the organization; discussion is useful when it connects to a real decision or change people can make.
Some organizations are still establishing basic policies and safe practices, while others are already reviewing the consequences of established use. This reinforces the importance of beginning by listening rather than assuming that every organization needs the same tool, training or intervention. Where data protection, consent or technical suitability is at issue, the relevant specialists and decision-makers need to be involved.
We are grateful to everyone who contributed to this inquiry. Their willingness to share their experiences, questions and reflections has not only informed this report but has also helped us think more deeply about how organizations can approach AI integration through collective learning, thoughtful experimentation and meaningful participation.
The thinking behind the approach
Experiential learning connects action with reflection and further testing. Morris's review emphasizes active involvement, real problems and context, while noting that the revised model needs further empirical testing. [10] Our proposed process applies that orientation to organizational work.
Schein's process consultation places the client's inquiry at the centre. [11] Senge's learning-organization perspective and Theory U inform our attention to connections, assumptions and learning before committing to solutions [12, 13].
For us, human systems intervention means helping people examine how their actions, relationships and organizational conditions affect one another. Awareness becomes useful when it informs a decision, a changed practice or a better question to test.
From findings to shared dialogue
We hope the next online gathering will be more than a presentation of results. We want to invite participants to challenge our interpretation, compare what has changed since their interviews and share practices that others might adapt. We will also ask whether an ongoing, participant-shaped community of practice would be useful, and what form of exchange people would actually value.
Join us in January 2027
We are developing an online conversation to compare experiences across organizations and test the interpretation in this report. We would like original participants to help identify what rings true, what needs correction and what has changed since we spoke.
The next round also needs people whose daily work is affected by AI, including those who have found useful applications and those who see limits or choose not to use it. Join HumLab in January 2027 for an interactive online webinar exploring the findings of Human-Centred AI Learning. Together, we will reflect on the research, exchange experiences across organizations, and explore the possibility of establishing an ongoing community of practice for leaders navigating human-centred AI integration.
Register for the webinar ↗HumLab also facilitates this kind of inquiry within organizations, connecting people's experience to practical decisions about learning and work. A useful first conversation begins with one situation you want to understand better.
How this report was developed
The material and its scope
The report draws on 14 supplied, edited interview transcripts with leaders and specialists across several organizational and professional settings. The transcripts were not checked against audio. Fourteen conversations do not mean fourteen independent organizations: some participants discussed several settings, and one conversation primarily explored a potential commercial partnership.
The material mixes direct experience, accounts of other people's experience, plans and forecasts. Interview quotations are anonymized, with source codes retained in an internal evidence register rather than shown in the public report. Quotations reproduce wording in the supplied edited text; they should not be understood as audio-verified verbatim quotations.
Survey context
Sections “Who took part” and “Survey results” reproduce descriptive material from the companion HumLab survey report. Fourteen people completed the survey; 14 were interviewed, with 13 participating in both, according to that report. The survey findings were not used to revise the interview evidence or the analytic method described below.
Analysis and the use of AI
This report was developed through a predominantly human-led process of inquiry, analysis, interpretation, and collaborative writing and editing, with AI used as a supporting tool. Our team conducted all 14 interviews, reviewed and reread the transcripts, and engaged in ongoing discussions to identify emerging patterns, explore differences in participants' experiences, and consider alternative interpretations.
AI was used to assist with organizing the material, identifying provisional patterns, comparing accounts, and assembling supporting and contrasting evidence. These outputs were critically reviewed, questioned, and revised by our team, who returned to the original transcripts to verify interpretations and ensure that the findings accurately reflected participants' experiences. The analysis, conclusions, and recommendations were developed and refined through human discussion and professional judgment.
Research and interpretation
The cited research was selected to explain, compare with or qualify issues raised by the interviews. This is a focused, non-systematic selection. It includes empirical studies, research reviews and conceptual approaches; adding these sources does not make the interview study representative or a mixed-methods evaluation. HumLab's recommendations are proposals informed by this material, not interventions tested in the study.
Limits that affect the conclusions
The sample mainly represents leaders and specialists. Employee views are often reported indirectly. Questions emphasized human and organizational issues, and some interviewer examples were repeated across conversations; those repetitions were not counted as independent evidence. The study does not estimate prevalence, establish saturation or determine causal effects on productivity, employment or wellbeing.
Our consulting interests inform the inquiry and interpretation. Participant feedback and broader involvement in a next round can challenge that interpretation. Identifiable transcripts and the detailed evidence register are not intended for public release.
Research on work, learning and governance
Numbers refer to research citations in the text. The notes identify what each source contributes and the limits of that connection.
Brynjolfsson, E., Li, D., & Raymond, L. (2025). Generative AI at work. The Quarterly Journal of Economics, 140(2), 889–942. doi.org/10.1093/qje/qjae044
Workplace study of customer-support agents. Supports bounded performance and learning claims; its effects are not estimates for our interview settings.
Brynjolfsson, E., Rock, D., & Syverson, C. (2021). The productivity J-curve: How intangibles complement general purpose technologies. American Economic Journal: Macroeconomics, 13(1), 333–372. doi.org/10.1257/mac.20180386
Economic model and analysis of complementary investment. Helps explain measurement and timing; does not validate any individual AI project.
Morrison, E. W. (2023). Employee voice and silence: Taking stock a decade later. Annual Review of Organizational Psychology and Organizational Behavior, 10, 79–107. doi.org/10.1146/annurev-orgpsych-120920-054654
Research review on speaking up and silence. Frames the inquiry into voice; employee influence in our cases remains an empirical question.
Lapointe, L., & Rivard, S. (2005). A multilevel model of resistance to information technology implementation. MIS Quarterly, 29(3), 461–491. doi.org/10.2307/25148692
Longitudinal case research in hospital information systems. Provides a comparison for resistance and power, rather than direct evidence about generative AI.
Blume, B. D., Ford, J. K., Baldwin, T. T., & Huang, J. L. (2010). Transfer of training: A meta-analytic review. Journal of Management, 36(4), 1065–1105. doi.org/10.1177/0149206309352880
Meta-analysis of 89 studies. Supports attention to learning conditions; varied tasks, measures and study designs limit specific prescriptions.
Vaccaro, M., Almaatouq, A., & Malone, T. (2024). When combinations of humans and AI are useful: A systematic review and meta-analysis. Nature Human Behaviour, 8, 2293–2303. doi.org/10.1038/s41562-024-02024-1
Experimental evidence across tasks. Distinguishes improvement over humans alone from outperforming the better standalone performer.
Beane, M. (2019). Shadow learning: Building robotic surgical skill when approved means fail. Administrative Science Quarterly, 64(1), 87–123. doi.org/10.1177/0001839217751692
Ethnographic and interview research on surgical learning. Raises a mechanism to investigate in other settings; it is not a study of office generative AI.
Rakova, B., Yang, J., Cramer, H., & Chowdhury, R. (2021). Where responsible AI meets reality: Practitioner perspectives on enablers for shifting organizational practices. Proceedings of the ACM on Human-Computer Interaction, 5(CSCW1), Article 7. doi.org/10.1145/3449081
Qualitative practitioner research. Connects responsible-AI work to organizational arrangements; does not settle meeting-consent or legal questions.
Parker, S. K., & Grote, G. (2022). Automation, algorithms, and beyond: Why work design matters more than ever in a digital world. Applied Psychology, 71(4), 1171–1204. doi.org/10.1111/apps.12241
Research-based conceptual analysis. Frames choices about job demands and resources; not a causal estimate of effects in our sample.
Morris, T. H. (2020). Experiential learning—a systematic review and revision of Kolb's model. Interactive Learning Environments, 28(8), 1064–1077. doi.org/10.1080/10494820.2019.1570279
Review and proposed learning model. Informs our process; the author explicitly calls for further empirical testing of the model.
Schein, E. H. (1995). Process consultation, action research and clinical inquiry: Are they the same? Journal of Managerial Psychology, 10(6), 14–19. doi.org/10.1108/02683949510093830
Conceptual account of consultation and inquiry. Informs the relationship with clients and how questions are developed.
Senge, P. M. (2006). The fifth discipline: The art and practice of the learning organization. Revised edition. Doubleday. penguinrandomhouse.com
Practice framework for organizational learning and systems thinking. Used as a guiding tradition rather than evidence for intervention effectiveness.
Presencing Institute. Theory U. Approach and practice resources. presencing.org/theoryu
Practice framework informing attention to assumptions, wider perspectives and experimentation. This is not an empirical evaluation of HumLab's approach.
How is AI changing work in your organization?
What are people learning, and which questions remain unresolved? HumLab helps organizations bring people together to understand their experience, learn from one another, and shape practical changes to how work gets done. Our approach draws on process consultation, human systems intervention and experiential learning.
Get in touch to discuss a challenge in your organization, express interest in our upcoming online conversation, or contribute to the next round of this inquiry.
If you participated in the interviews, we would especially welcome your response: what feels accurate, what needs correcting, and what has changed since we spoke.
- Emailteam@humlab.ca
- LinkedInlinkedin.com/company/humlab-consulting
- Websitewww.humlab.ca
- WebinarRegister for January 2027 ↗