1. Why this case matters

Much of the discussion about AI focuses on technology, productivity, policy or future scenarios.

Will Allen takes a different starting point: everyday practice.

The article reflects on what may be changing as AI enters evaluation, facilitation and collaborative decision-making in place-based settings.

These settings can involve:

• catchments and regions;
• freshwater management;
• land use;
• community change;
• environmental issues;
• cross-organisational collaboration.

These are not abstract problems. They involve real places, histories, relationships, different interests and long-term consequences.

2. What is shifting?

Allen uses a futures-thinking lens, but importantly, not as a prediction exercise.

It is a way of noticing movement in practice:

• what ways of working may be receding;
• what pressures are becoming more prominent;
• what new possibilities are appearing;
• what new risks are beginning to emerge.

AI may not arrive through one dramatic organisational transformation. It can enter quietly through everyday activities such as preparing, drafting, analysing and reflecting.

3. AI is becoming less peripheral

One of the shifts identified in the article is that the assumption that AI is simply optional or sits at the edge of professional practice is becoming less certain.

People may already be using AI to prepare, draft and reflect, sometimes without this being formally recognised by their organisation.

This creates an important distinction:

Formal AI adoption — an organisation deliberately introduces an AI system.

Informal AI adoption — staff begin using AI in everyday work before organisational norms or guidance are fully developed.

For evaluators, the second situation matters too. AI use can become part of the organisational context that an evaluation needs to understand.

4. Speed and polish can change how people work

AI can produce fluent early drafts quickly. This can be useful, particularly during the early stages of thinking.

But Allen also points to a less obvious consequence: the rhythm of professional work may change.

Previously, professionals might have spent time working through incomplete information, drafting rough ideas, reworking them and gradually developing judgement.

With AI, more of that initial material may arrive already structured and polished.

5. Why place-based work is different

Place-based work often involves different values, knowledge and priorities.

For example, people working on a catchment or land-use issue may disagree about:

• what the problem is;
• which evidence matters;
• what should be prioritised;
• what constitutes progress;
• which trade-offs are acceptable.

These differences are not necessarily something an evaluator should eliminate.

They may be part of the evidence.

6. The hidden learning question

Another issue raised by Allen concerns early-career learning and professional judgement.

Professionals often learn by working through messy material themselves: reading, drafting, revising, questioning and making sense of place-specific information.

If more of this early work becomes AI-assisted, an important question emerges:

How will professional judgement develop over time?

The article does not present this as an established negative effect. It is an emerging uncertainty that is worth watching.

7. Funders and commissioners are also part of the shift

Allen notes early signals that some funders and commissioners are beginning to ask about AI use or assume a level of AI-assisted work, even where norms are not yet well defined.

This can leave teams working out, in real time:

• when AI should be used;
• when it should not be used;
• how its use should be disclosed;
• who remains responsible for the output;
• how AI-assisted work should be reviewed.

For evaluators, this makes AI use relevant not only to analysis but also to commissioning, quality assurance and professional accountability.

8. Authorship, accountability and transparency

When AI contributes to shared documents or collaborative processes, questions of authorship and accountability become more visible.

9. What remains uncertain?

The article is valuable because it does not present these questions as settled conclusions.

Collective judgement: If participants use AI to prepare their inputs, will this broaden ideas or create greater convergence in how issues are framed?

Relationships: What happens to trust, timing and presence when some aspects of preparation and synthesis become automated?

Organisational culture: Will organisations develop deliberate approaches to AI, or move primarily because of pressure for efficiency and output?

Diversity of perspectives: Will AI smooth differences, or can it also help surface new connections and wider perspectives?

Professional expectations: If AI-assisted drafting becomes normal, will expectations of what constitutes competent or thorough work change?

Allen presents these as emerging questions rather than established effects.

10. Five practices organisations are beginning to work through

The article identifies five areas where useful practice is beginning to take shape. These are not presented as settled rules.

1. Make AI use visible. Be clear about when and how AI contributed to drafts, analysis or synthesis.
2. Keep responsibility for judgement with people. AI can support thinking, but people remain responsible for interpretation, evidence and decisions.
3. Check and ground outputs. Test AI-generated material against context, evidence and local knowledge.
4. Be selective about where it is used. Sensitive relationships, local knowledge and contested issues may require additional care.
5. Support staff and shared learning. Give teams space to experiment, reflect and develop shared understanding.

11. What should evaluators hold onto?

12. A practical example for evaluators

Scenario: evaluating a climate-resilience programme

An international development organisation is evaluating a five-year climate-resilience programme operating across several rural catchments.

The evaluation team has community interviews, farmer consultations, stakeholder meeting notes, government documents, partner reports and environmental monitoring information.

The team wants to use AI to help organise and synthesise the qualitative material.

Step 1 — Use AI to organise, not decide

Instead of asking:

the evaluator could ask:

The objective is to structure evidence without asking AI to determine the evaluation’s interpretation.

Step 2 — Check what the summary has compressed

Suppose the AI produces:

The evaluator returns to the original evidence and discovers different levels of support across stakeholder groups.

The AI statement may not be entirely false, but it may have smoothed important differences.

Step 3 — Apply an AI perspective check

Ask:

• Who is represented?
• Whose perspective may be missing?
• What disagreement remains?
• Are minority views visible?
• What contextual information disappeared?
• Which statements can be traced to the original evidence?
• What requires human judgement?

Step 4 — Use AI to challenge the evaluation team

AI does not have to produce the final finding. It can instead help generate questions:

The evaluator then returns to the evidence, investigates the questions and, where appropriate, engages directly with stakeholders.

Step 5 — Ask what changed because AI was used

This moves the focus from AI accuracy alone to the quality of the evaluation process itself.

13. A practical daily workflow

1. Define: What exactly are we asking AI to do?
2. Organise: Use AI to structure complex information.
3. Preserve: Keep disagreement, minority perspectives and uncertainty visible.
4. Verify: Check important outputs against original evidence.
5. Interpret: Apply professional and contextual judgement.
6. Challenge: Ask AI to identify gaps, assumptions and alternative interpretations.
7. Engage: Return to people and stakeholders where context matters.
8. Document: Make AI use and human responsibility visible.

14. What should evaluators monitor?

AI output: Is the output traceable, accurate and appropriately qualified?

Perspectives: Are minority, local and lived perspectives still visible?

Disagreement: Has disagreement been preserved or prematurely smoothed?

Professional judgement: Are people still interpreting and challenging evidence?

Relationships: Has AI changed how people listen, engage and work through difficult issues?

Learning: Are professionals developing judgement or becoming increasingly dependent on AI-generated first drafts?

15. Five questions to use tomorrow

16. The bigger lesson for M&E and international development

The article does not argue that AI should be rejected. Nor does it claim that AI will inevitably damage place-based practice.

Instead, it invites practitioners to notice what is changing while remaining attentive to what still matters.

For evaluators and development professionals, AI is therefore not only a technology issue. It can also affect:

• how evidence is interpreted;
• how perspectives are represented;
• how professional judgement develops;
• how teams learn;
• how stakeholders participate;
• how decisions are shaped.

17. Evidence boundary

This case study is based primarily on Will Allen’s AI in place-based practice: what is shifting, published by Learning for Sustainability on April 3, 2026.

The source is a practice-based reflection, not a report of a specific AI intervention, programme or impact evaluation. It does not provide measured outcomes from a named AI implementation.

The discussion of emerging shifts, speed and polish, authorship and accountability, early-career learning, funder and commissioner expectations, collective judgement, relationships, organisational culture, diversity of perspectives and organisational responses is based on the author’s reflections.

The daily workflow, evaluator questions, perspective check and climate-resilience example are EvalCommunity Academy’s educational interpretation and application. They are not presented as findings from the original article.

EvalCommunity Academy
Practical knowledge for evaluators and development professionals.