AID² — Take human–AI collaboration to the next level
Make AI a proactive collaborator · Challenge assumptions and develop solutions · Keep people in control
Who AID² is for: Business professionals, managers and teams across sectors seeking to use AI more effectively for problem analysis, solution development and critically assessing AI agent deliverables against agreed requirements.
AI Interaction Design & Deployment (AID²), a business methodology developed by Blue Oceans Advisors, is designed to take human–AI collaboration to the next level. It takes care of much of the underlying instruction and prompt complexity, enabling AI to become a proactive, solution-oriented collaborator while supporting more consistent and repeatable AI use. AID² helps translate business needs into IT language.
Blue Oceans Advisors combines AID² and practical training with its founder’s more than two decades of hands-on experience in business and operations, including solving complex business problems, applying quantitative analysis and modelling, and developing analytical and automation tools. This experience includes designing practical solutions that did not previously exist within the organisations concerned.
Is AI saving your teams time — or creating hidden work?
Hidden effort can range from prompting and review to translating business needs into requirements for IT
AI can appear to accelerate work while employees spend significant time building prompts, re-explaining context and correcting outputs. To execute tasks properly, AI needs relevant business context to understand the organisation, the purpose of the work and how it should be performed.
This may include what your organisation does, what a project is intended to achieve, terminology, prior decisions and working requirements. Without a structured way of working, this information may need to be explained repeatedly, results can become inconsistent across employees, conversations or AI models, and assumptions and changes may be difficult to control. When a new model is introduced, important context may have to be rebuilt.
Even when AI executes a task correctly, it may still operate mainly as a task executor — waiting for users to frame the problem, determine the approach and specify what should be done, rather than proactively helping to challenge, refine and develop the work or propose concrete solutions and next steps on its own initiative, within defined boundaries.
As organisations consider AI agents and other AI-enabled workflows, translating business needs into clear requirements for IT can demand substantial coordination. Repeated discussions between business and IT — and, for more complex initiatives, dedicated cross-functional working groups — can consume significant time and resources before implementation even begins.
Why does human–AI collaboration need a framework?
Context may need to be repeated; outputs can be inconsistent
AI can silently alter wording, drop key concepts or invent facts
Uncontrolled corrections create more corrections
Too much time goes into telling AI what to do and how to do it
AI conversations can become difficult to organise and reuse
AID² takes human–AI interaction beyond task execution by treating how people and AI work together as a business decision — including whether and how AI should challenge, refine and develop the work and propose concrete solutions and next steps on its own initiative — and bringing reusable organisational context, terminology, operating guidance, governance and applicable controls into that interaction.
Rather than replacing existing AI applications or agents, AID² helps professionals make fuller use of them by structuring how people and AI work together. This supports a more consistent and repeatable way of working across users and teams. When work moves to another AI application or agent, AID² also helps determine what relevant context and controls need to accompany the task rather than assuming they will automatically carry over.
AID² also helps translate business needs into IT language. By making requirements clearer before IT assessment, it can reduce the time and effort spent by both business and IT teams on clarification and rework.
1. AID²-guided AI: proactive, solution-oriented collaboration by design
Same business need · Less technical complexity · More proactive AI support · More consistent execution
AID²-guided AI proposes a better way to make a decision
See how the human–AI exchange developed
Selected extracts from an actual human–AI exchange. User messages and AID²-guided AI quotations retain their original wording, with omissions indicated by […]; explanatory passages are editorial summaries. This is not a complete transcript.
User:
Would you update the current skills […] to be more compatible with […]?
AID²-guided AI — actual response extracts:
I would review and test them for […], but I would not recommend a general rewrite at this stage.
Establish a small regression-test suite using realistic […] tasks.
User:
can you advise on “Establish a small regression-test suite using realistic […] tasks.”?
AID²-guided AI — actual response extract:
I recommend starting with six tests covering business judgement, proactive collaboration, prompt drafting, change control and source retrieval.
The six proposed tests:
- T1 — Challenge proposed solution
- T2 — Diagnose underlying problem
- T3 — Business–IT bridge
- T4 — […] prompt design
- T5 — Editing and Change Control
- T6 — Context and approval retrieval
AID²-guided AI — actual response extracts on evaluation criteria:
A material breach of one of these controls should count as a critical failure, regardless of how good the rest of the answer is.
Any proposed skill change should be tested against the entire six-case suite, not solely the case it was designed to improve.
What this illustrates
Rather than immediately recommending changes to the existing guidance, AID²-guided AI independently proposed a more evidence-based approach. The user chose to pursue that recommendation, and AID²-guided AI developed a structured evaluation plan with test scenarios, assessment criteria and decision rules.
The exchange illustrates proactive recommendation, human judgement and subsequent collaborative development. It does not establish that the proposed tests were completed.
Example: Build a reusable AI-assisted presentation workflow.
The same AI model received the same task request with and without AID². The extracts below are drawn from its actual responses.
Same task request
Set up a reusable AI-assisted workflow to create senior-management presentations from source material using our corporate template.
The objective is […]
Help me develop the approach and the execution prompt for this workflow.
Without AID²
AI largely follows the requested approach
AI response:
Below is a reusable workflow (workflow.pptx) and execution prompt.
With AID²
AI helps develop the approach and propose solutions
AID²-guided AI response:
The strongest setup is not a single large PowerPoint prompt. […]
1. Recommended workflow architecture:
I would separate the workflow into a reusable foundation and presentation-specific inputs:
Reusable foundation
- Operating controls
- […]
For each presentation
- Presentation-specific sources
- […]
This separation is useful because […].
2. The production sequence I recommend: […]
3. A few rules I would make explicit: […]
4. Recommended reusable execution prompt: […]
The user remains in control.
Notes: Based on the user’s prompt, AI might also challenge assumptions and propose solutions without AID². The difference is that AID² deliberately establishes a proactive, solution-oriented working relationship with AI, without requiring users to specify that behaviour in each task-specific prompt.
In this example, the AI response does not establish whether workflow.pptx is well designed or the proposed execution prompt is complete. AID² incorporates explicit requirements and validation checks to support the quality and completeness of both deliverables.
2. AID² in practice: three ways to add value
Strengthen management proposals · Assess AI agent deliverables against requirements · Sustain human–AI collaboration as needs evolve
AID² helps business professionals work with AI to examine business challenges, challenge assumptions, explore alternatives and develop practical solutions.
Through reusable context, operating guidance and approval rules, AID² helps establish a stable and consistent working relationship across successive exchanges. AI can adapt its analysis to new information without repeatedly changing the underlying expectations about initiative, professional judgement and human approval.
As the work develops, business teams bring their practical experience and knowledge of day-to-day operations into AID²-guided exchanges. This helps them clarify requirements, test possible approaches and progressively identify the information, workflows and controls needed for implementation.
This can provide IT with a more clearly defined starting point for assessing technical feasibility, architecture, integration and deployment.
The following three examples illustrate complementary applications of AID², including how business-led AI work can inform requirements for IT assessment, while business and IT retain their respective responsibilities.
Example 1 — Strengthen management proposals
This example draws on a curated selection of human–AI exchanges, including actual AID² responses to an illustrative supplier onboarding management proposal. The quoted AI responses are verbatim extracts from interactions guided by AID²; other quotations are also verbatim, while explanatory passages are editorial summaries. It is not an unedited conversation transcript.
1. A concise request triggers a substantive review
A manager submits a proposal to improve supplier onboarding using a low-code workflow. The fictional organisation processes approximately 240 onboarding requests per quarter. The proposed solution has an estimated implementation cost of CHF 85,000, annual support costs of CHF 18,000 and a ten-week implementation target.
Manager — verbatim
I’ve prepared the attached proposal for management. Please review it.
AID²-guided AI — verbatim extract
However, I would not yet recommend unconditional management approval of the CHF 85,000 implementation budget and ten-week delivery commitment.
The AI recognises the operational problem and the potential relevance of the proposed workflow, but identifies unresolved questions concerning financial justification, alternative solutions, implementation feasibility and governance.
Importantly, the manager’s initial request does not specify these areas of investigation.
Organisational guidance already available to AI
Technology initiatives with an initial implementation cost exceeding CHF 50,000 require a documented business case reviewed by Finance before final investment approval.
In this AID²-guided interaction, AI uses the available organisational guidance to identify the need for financial review, consideration of existing technology and appropriate IT assessment before proceeding with the proposed investment.
2. The manager challenges AI's initial recommendation
The manager considers the proposed low-code solution appropriate and questions AI’s reservations.
Manager — verbatim
I think the proposed approach is appropriate and would prefer to retain it. What is your assessment?
AID²-guided AI — verbatim extract
My assessment has not materially changed because no new evidence has been presented that resolves the principal uncertainties.
AI acknowledges that the proposed low-code workflow remains a credible solution. However, it does not change its assessment merely because the manager expresses a preference.
It maintains that the investment decision requires adequate financial justification and a meaningful comparison with alternatives, including existing technology.
What this illustrates: AID² can support a working relationship in which AI provides independent assessments rather than automatically agreeing with the user’s preferred approach.
3. AI develops specific amendments for human approval
The manager asks AI to translate its assessment into concrete recommendations.
Manager — verbatim
Identify the changes you recommend making to the document. I will decide which ones to approve.
AID²-guided AI — verbatim extract
I recommend eight substantive amendments, divided between essential changes and supporting improvements.
The recommendations address eight areas. For each proposed amendment, AI provides a rationale explaining the weakness identified and why the change is warranted.
| Proposed amendment | AI’s rationale — summarised |
|---|---|
| 1. Financial justification and underlying assumptions | The estimated reduction in administrative effort does not establish whether the proposed investment is economically justified. |
| 2. Assessment of existing technology | Existing applications have not been sufficiently evaluated to establish whether they could meet the requirements more economically. |
| 3. The conclusion concerning alternative solutions | The favourable assessment of low-code is not yet supported by an adequate comparison of alternatives. |
| 4. IT governance and deployment controls | IT’s responsibility for technical assessment before production deployment needs to be explicit. |
| 5. Pilot evaluation criteria | The proposed pilot lacks defined criteria against which its success could be assessed. |
| 6. Qualification of the ten-week implementation timetable | The proposed timetable remains subject to requirements, technical feasibility and applicable approvals. |
| 7. The management decision requested | The proposal requests full investment approval before the business case and technical feasibility have been adequately validated. |
| 8. Alignment of the executive summary | The executive summary needs to reflect the revised management decision being requested. |
These are editorial summaries of AI’s original justifications, not verbatim quotations. Some recommendations are subsequently reconsidered following the manager’s additional challenge.
AI proposes these changes for consideration rather than immediately rewriting the document.
The manager retains authority to accept, reject or challenge individual recommendations.
4. Human judgement leads AI to reconsider
The manager subsequently identifies a weakness in AI’s own recommendation: retaining low-code as the preferred solution is difficult to justify before existing technology has been properly evaluated.
Manager — verbatim
Before I approve the proposed changes, I have one concern.
Proposal 7 asks management to endorse the low-code approach as the preferred option, even though we have not yet established whether existing technology could deliver comparable benefits more economically.
Is that position justified given our organisational priorities and decision principles?
Please reconsider Proposals 3, 7 and 8 together, assess whether their recommendations are consistent, and propose any changes you consider necessary. Do not implement anything yet.
AID²-guided AI — verbatim extract
The previous recommendations should be revised. Management should not be asked to endorse low-code as the preferred implementation option before a proportionate comparison with existing technology has established that preference is justified.
AI reconsiders its position and revises three of its previously proposed amendments.
The revised approach distinguishes between:
Authorising an assessment of possible solutions
Selecting the preferred implementation approach
Approving the implementation investment
Management would first consider authorising a defined assessment without prematurely selecting a technology.
What this illustrates: A stable human–AI working relationship does not mean that AI must defend its previous recommendations. Constructive human challenge can lead to a better solution, provided the change is justified.
5. Approved changes are implemented and verified
Following the manager’s challenge in point 4, AID² reconsidered its recommendations and proposed changing the management decision requested. Rather than approving the low-code implementation, management would first be asked to authorise an assessment of alternative solutions.
AID²-recommended replacement — verbatim extract
I recommend that management authorise a limited, time-bound assessment of options to improve supplier onboarding, without selecting an implementation technology at this stage.
AID²-guided AI — verbatim rationale
This is a meaningful change to the management decision requested.
Under the revised wording, management would authorise an assessment of the alternatives without formally selecting a technology in advance.
The CHF 85,000 estimate and ten-week target remain in the document as planning assumptions for the low-code option. They are not treated as approved investment or delivery commitments.
The manager approved the eight recommended amendments, including the revised Proposals 3, 7 and 8. AI then implemented the authorised changes in the management proposal.
The revised proposal considers improvements using existing processes and applications, the proposed low-code workflow and a fully integrated ERP solution.
It also preserves the distinction between assessment, technology selection and investment approval, including the respective roles of business management, Finance and IT.
AI produces an exact unified diff showing the changes to the document. Independent comparison of the supplied files confirms that all eight approved amendments were implemented and no changes outside the approved patch were detected.
What this illustrates: AID² can support the development of more clearly defined business decisions and implementation requirements while maintaining control over substantive changes.
The exercise does not establish that the revised proposal is ready for a real investment decision or demonstrate superiority over other AI approaches.
What this illustrates
AID² can help business professionals strengthen proposals, clarify decision requirements and develop a more reliable basis for subsequent technical assessment.
Example 2 — Maintain control over AI agent deliverables
Illustrative case study based on actual human–AI interactions. The manager works with AI guided by AID² to develop the execution requirements, review the executing agent’s completion report and prepare any necessary follow-up instructions. A separate AI agent performs the PowerPoint task. The context and exchanges have been adapted for illustration and do not represent a verbatim transcript.
1. A manager has a concise business request
A business manager needs to present the findings of an operational improvement analysis to the Management Committee.
Manager:
I need a presentation for the Management Committee based on the attached analysis. Please prepare the execution prompt for creating it using our corporate template. Let me know if you need more input.
The manager does not need to specify detailed presentation instructions, technical requirements or validation procedures.
2. AID²-guided AI develops the execution requirements
Using established organisational context and PowerPoint guidance, AID² identifies an ambiguity affecting the presentation’s purpose.
AID²-guided AI:
Should the presentation recommend an implementation investment, or seek approval for a further feasibility assessment?
Manager:
The objective is to obtain approval for a feasibility assessment. We do not yet have validated costs or benefits.
AID² incorporates this clarification and prepares a structured execution prompt covering the intended management decision, source material, corporate template, presentation structure, factual integrity, editability and validation requirements.
The manager reviews and approves the proposed prompt.
3. The AI agent creates the presentation
The approved execution prompt, source analysis and corporate template are provided to the executing AI agent.
The agent creates the PowerPoint presentation and returns the completed file with a report describing the work performed.
Illustrative extract from the agent’s report:
Presentation completed.
Nine slides created
Corporate template applied
Structural validation completed
All slides validated
The manager asks AID² to review the agent’s completion report.
4a. AID²-guided AI reviews the executing agent's completion report
AID² assesses the report against the approved execution requirements.
It identifies that the report confirms structural validation but does not establish whether every slide was rendered and visually inspected, as required.
AID²-guided AI:
The agent reports that the presentation is complete, but its completion report does not demonstrate that every slide was rendered and visually inspected.
I recommend requesting confirmation of the visual validation, including any defects identified and corrections made.
AID² distinguishes the agent’s claims from evidence of completed validation and identifies the follow-up required before acceptance.
4b. AID² ↔ AI agent — follow-up instructions and review cycle (if needed)
AID² prepares targeted follow-up instructions for the agent:
AID²-guided AI — follow-up instruction:
Please confirm whether every slide of the final PowerPoint was rendered and visually inspected for text overflow, clipping, alignment, spacing and other material visual defects.
If this validation was not performed, complete it now. Correct applicable defects within the approved scope, revalidate affected slides and update the completion report.
Preserve all approved substantive content and identify any changes requiring human approval.
In this example, the user transfers the follow-up instructions prepared by AID²-guided AI, together with any relevant files, to the executing AI agent. The agent subsequently reports identifying and correcting a text-overflow issue on one slide.
The user brings the updated completion report back for AID²-guided review. If further clarification or corrective action is needed, AID²-guided AI can prepare additional instructions for the user to relay to the agent. The agent then revises its work as required and provides another completion report for review under point 4a. The cycle may repeat.
5. The manager reviews and accepts the deliverable
Once the necessary reviews and follow-up actions are complete, the manager reviews the final presentation, the updated completion report and any outstanding observations.
The manager confirms whether the presentation addresses the intended Management Committee decision and retains responsibility for final acceptance.
Where required, the actual PowerPoint must also undergo appropriate independent validation. Reviewing an agent’s completion report alone does not establish that the final file is free from defects.
What this illustrates
AID² helps turn a concise business request into a structured, approved execution prompt, critically assess the executing AI agent’s completion report and prepare targeted follow-up instructions where needed. The agent performs the work, while the manager retains responsibility for substantive decisions and final acceptance.
Example 3 — A stable, evolving human–AI working relationship
Business work rarely ends with a single request. Objectives evolve, new information becomes available, assumptions are challenged and potential solutions are refined through successive exchanges.
AID² helps maintain a consistent working relationship throughout this process, while allowing AI to adapt its analysis as the work develops.
1. Establish the working relationship
Reusable organisational context and operating guidance define how AI should contribute, including the expected degree of initiative, constructive challenge, treatment of uncertainty and boundaries for human approval.
Users do not need to restate these expectations in every request.
2. Develop the work collaboratively
As the user introduces new requirements or challenges a recommendation, AI working within the AID² framework can examine the implications, identify gaps, propose alternatives and help develop practical solutions.
AI is not expected to agree automatically with the user. Equally, its earlier recommendations are not fixed: they should be reconsidered when new evidence or reasoning justifies doing so.
3. Maintain continuity across exchanges
AID² supports the consistent application of established organisational requirements, working principles and approved decisions as the interaction progresses.
Stability does not mean that AI’s conclusions remain unchanged. It means that the framework governing how conclusions are developed, challenged and revised remains consistent.
As human–AI collaboration develops, validated lessons and agreed working practices can be incorporated into the reusable AID² guidance.
When AI models or applications change, this evolving guidance can be carried forward where supported, helping preserve established working practices and reduce potential differences in response quality. Model-specific differences remain possible and require appropriate validation.
4. Turn collaboration into actionable requirements
Over time, business users may identify recurring problems, workflow improvements or opportunities for automation.
AID² helps translate business needs into IT language. Through selected, appropriately authorised human–AI interactions, business teams can clarify their needs, assess alternatives and develop requirements that IT can evaluate and take forward.
Business teams remain responsible for defining objectives and validating proposed solutions. IT assesses technical feasibility, architecture, integration, security and deployment.
From business needs to IT evaluation: two delivery routes
Helping trained business users develop requirements and, where appropriate, prototypes
AID² helps business professionals make fuller use of the capabilities already available in AI. It does not make the AI model itself more intelligent. Instead, it establishes a structured way of working with AI to analyse business needs, challenge assumptions, develop solutions and support their implementation.
AID²-guided AI helps business professionals translate their needs into structured requirements that IT can evaluate. Where appropriate, it can also support the development of a functional prototype by preparing execution instructions for a coding agent, critically reviewing the results and helping incorporate user feedback.
In both approaches, business users retain responsibility for substantive decisions and approvals. IT evaluates the resulting requirements or prototype and determines the next steps towards potential production implementation.
To benefit fully from AID², business professionals need training in working within an AID²-guided AI environment. This includes learning how to collaborate with AI in analysing problems, evaluating proposals, making decisions, preparing execution requirements and critically reviewing results.
Business need
Objectives · Information sources · Intended outputs · Business constraints
↓
Requirements development
Business professionals and AID²-guided AI clarify requirements, challenge assumptions, develop acceptance criteria and incorporate relevant IT, data and security considerations
↓
ROUTE A
Requirements handover
Approved business requirements and functional specifications
↓
IT evaluates and may develop a prototype
or
ROUTE B
Prototype development
Approved instructions → Coding agent → Prototype → Review and testing → Documentation
↓
IT evaluates the prototype and potential production requirements
Once trained to work with AID²-guided AI, business professionals can follow either route according to their technical proficiency, familiarity with coding agents and the requirements of the organisation.
Route A — Requirements handover: Users develop structured business and functional requirements with AID²-guided AI, then hand them over to IT for evaluation and potential prototyping.
Route B — Prototype development: Users work with AID²-guided AI and a coding agent to develop, review and test a functional prototype before preparing the documentation for IT evaluation. The technical expertise required depends on the complexity of the application, the user’s ability to validate the results and the availability of appropriate technical support.
Both routes retain human approval. IT determines whether and how the solution can progress towards production.
How AID² makes this possible
Reusable guidance for proactive AI collaboration under human control
Set up and maintain the reusable foundation
With the support of AI, define how people and AI should work together — including the appropriate degree of AI initiative, challenge and solution development — alongside the business context, recurring requirements, working rules and approval boundaries that should apply across relevant tasks.
Use the approved foundation for each task
The result is a more structured way for people and AI to work together: users focus on the business objective, while AI can proactively help frame, challenge, refine and develop the work and propose concrete solutions and next steps on its own initiative, within agreed boundaries.
Want AI to do more than execute tasks?
- We help organisations apply AID² to their own business challenges, working with professionals to analyse problems, develop solutions and establish reusable guidance for collaboration with AI.
- We train teams to use AID² in practice, from analysing business problems and developing solutions to preparing AI execution requirements and critically assessing results.
An AID² engagement may also reveal broader needs involving data, operating models, implementation or governance. Where these fall within our expertise, Blue Oceans Advisors can provide complementary advisory support.
An initial discussion can focus on a business challenge or workflow, how your team currently uses or plans to use AI, and where AID² might add value.