From ChatGPT to AI Agents, AI Loops and AI Graphs: the next evolution of AI in management consulting
AI is moving beyond answering questions. Agents execute tasks, loops refine results, and graphs connect intelligence across whole workflows. At Wang Advisory I have turned that evolution into a working toolkit for management consulting, and this is how the backbone fits together.
AI is no longer the thing you ask, it is the thing that runs the work.
ChatGPT changed how we interact with artificial intelligence. AI agents are changing how we execute work. AI loops introduce iteration and refinement. And AI graphs connect agents, tools, data and decisions into one system.
For management consulting this is more than a technology trend. It changes how we structure problems, analyse information, run due diligence and deliver executive-ready recommendations. It also changes what a consulting firm has to be good at: not prompting, but designing the workflow and owning the judgement at the end of it.
The four phases of AI evolution
Phase 1: ChatGPT, AI that answers
The first wave made AI accessible to almost everyone. We could ask questions, summarise documents, brainstorm and generate content in seconds. But the user still had to define the workflow, connect the information, validate the output and turn it into something useful.
The breakthrough: AI made knowledge more accessible.
Phase 2: AI agents, AI that executes
Agents take the next step. Rather than responding to a single prompt, they follow defined instructions, use tools and carry multi-step work towards a specific objective. In consulting that means research, problem structuring, financial analysis, due diligence and presentation development.
The breakthrough: AI moves from generating answers to completing tasks.
Phase 3: AI loops, AI that iterates and improves
A loop is a cycle in which AI plans, executes, evaluates the result and uses that feedback to decide what to do next. Instead of producing an output once and stopping, the workflow checks its work against defined criteria, identifies gaps, refines the result and repeats where appropriate.
The breakthrough: AI workflows become iterative rather than one-shot.
Phase 4: AI graphs, connected intelligence
Graphs connect specialised agents, tools, data sources and workflow dependencies into a coordinated system. One agent's output becomes another agent's input, while the workflow coordinates dependencies and routes tasks where they are needed. Graphs also support traceability, defined handovers and human approval points, which matter in M&A and due diligence.
The breakthrough: AI capabilities become connected into coordinated workflows.
These four phases are a framework for understanding the evolution, not a universally agreed industry classification. In practice agents, loops and graph-based workflows overlap.
The loop, not the prompt, is where quality is made
The difference between a demo and a deliverable is the second pass. A one-shot answer stops whether it is right or wrong. A loop sets the criteria before it starts, tests its own output against them, names what is still missing and runs again.
Plan: set the hypothesis, the criteria the answer must meet and the evidence it needs.
Execute: run the analysis, draft the storyline, build the exhibit.
Evaluate: test the output against those criteria and name the gaps that remain.
Refine: close the gaps, raise the open question for a human, and run the pass again.
Loops do not guarantee correctness. Their value depends on the quality of the data, the evaluation criteria, the tools and the human oversight around them. That is why the standard is set by a person, not by the model.
The toolkit: one core agent, specialists on demand, one delivery backbone
This is the toolkit I deployed with the launch of the Wang Advisory Digital Store. It is not a collection of separate tools, it is a backbone with three layers and a human on top of all of them.
The core: my daily consulting agent
The Management Consultant AI Agent is the one I work with every day. It structures the problem, runs the research and analysis, and turns findings into a recommendation across the full consulting toolkit, from strategy and value creation to transactions and transformation. Everything else plugs into it.
The deep dives: specialist agents, built when a mandate needs them
When a mandate calls for a specific capability, I build a specialist agent for it. The Financial Due Diligence Agent and the Technology Due Diligence Agent are the first two examples: one works through earnings quality, working capital and net debt, the other through applications, infrastructure, cyber and licensing. They are exemplary rather than exhaustive. The pattern is the point, and further specialists follow the same shape.
The backbone: the PowerPoint agent
Every engagement ends in a document. The PowerPoint Agent is the backbone that carries the work out of the analysis and into the room: storyline, action titles, exhibits, decks and Excel, in the firm's own design. It is the piece I need on every single project, which is why it sits behind all the others.
The layer that does not move: senior human review
At every handover a senior consultant approves the hypotheses, challenges the findings, checks every number against its source and owns the recommendation. AI-first does not mean AI-only. The agents carry the work; the consultant sets the standard and owns the call.
What the agents actually hand over
The test of an AI workflow is not the conversation, it is the artefact at the end of it. EBITDA and net debt bridges, adjustment breakdowns, databook analysis, peer benchmarks, cohort and cost structure views, findings and red-flag registers, executive summaries, dashboards and roadmaps. Every exhibit carries its units, its period and its source, and is checked before it reaches the room.
What is in the Digital Store
The store brings the toolkit together as digital consulting products, built as Claude Agent Skills:
Wang Advisory Management Consultant Claude AI Agent Skill: from an open-ended business problem to structured analysis, hypotheses and recommendations.
Wang Advisory Management Consulting PowerPoint Claude Agent Skill: from raw analysis to a structured, executive-ready presentation.
Wang Advisory Financial Due Diligence Claude Agent Skill: from financial data to focused diligence questions on revenue quality, margins, working capital, earnings and net debt.
Wang Advisory Technology Due Diligence Claude Agent Skill: from technology documentation to a structured view of risks, gaps and the questions to put to management.
Custom Claude PowerPoint Agent Skill in your corporate design: the same backbone, rebuilt in your firm's own brand.
The real opportunity: AI that fits the workflow
The future of AI in management consulting will not be defined by who writes the most sophisticated prompt. It will be defined by who builds the most useful workflows: specialised agents, iterative loops, connected handovers, reliable information, quality controls and human oversight.
AI can accelerate research, structure analysis and carry deliverable creation. Consultants still validate findings, challenge assumptions, apply judgement and take responsibility for the recommendation. In high-stakes work such as M&A and due diligence, an overlooked assumption or an unsupported conclusion has consequences.
The winning model is not human versus AI. It is human expertise amplified by well-designed, connected AI workflows. That is what AI-first means at Wang Advisory, and that is what I am building.
Explore the store
The four agent skills, and the custom PowerPoint agent in your own corporate design, are available in the Wang Advisory Digital Store.
Comments