Michelle Caryn Paul, PhD’s Writing Portfolio
– Please note that my prior name was Michelle Paul Heelan –
LinkedIn Posts
++++
The constraint used to be: How fast can we build a new service offering or product prototype?
A recent Fortune article and AJ Bubb‘s commentary makes the case that this constraint is gone. Vibe coding recently collapsed build time from weeks to hours.
So the bottleneck moved. For teams deploying AI and automation inside their organizations, the new bottleneck is: Do we understand our work well enough to redesign it?
Most teams do not. Not because they lack technical skill. Because no one has mapped how work actually flows before integrating AI or automation. The key question is no longer “What is the minimum we can build to introduce this new offering?”
The question is “What problem are we actually solving, and for whom?”
Getting those answers requires talking to the people doing the work, not just the people sponsoring the project.
Everyone can automate now. Those organizations that succeed are those that know precisely why they are automating.
Curious about how to map your workflows? Message me and I will send a one-page workflow mapping guide that can help you inventory your task, ownership, failure points, and where AI realistically fits.
++++
Most conversations about AI in the workplace contain some level of confusion. Someone says “context window” or “RAG” and half the group nods without knowing what the term means.
Most of us can name AI tools. Far fewer can explain what “tokens” or “temperature” actually means, or how those terms affect AI adoption decisions. Shreyas Naphad’s recent piece in Towards AI, Inc. features 5 important AI terms. Here is what each one means, and how each one can sharpen your decisions:
✅ Tokens: AI processes language in chunks. Knowing this helps you evaluate cost, set output expectations, and avoid overpaying for capability you are not likely to use.
✅ Context window: AI has a memory limit per conversation. Knowing this tells you which workflows are likely to break down and need redesign before you integrate AI.
✅ Temperature: This concept controls how consistent vs. unpredictable the AI output is. Knowing this helps you write usage policies that guide employees to match temperature to the task (e.g., set more reliability for compliance work, more flexibility for brainstorming).
✅ Hallucination: AI generates confident wrong answers. Knowing this helps direct your audits of output, where human review is non-negotiable, and what to put in your governance framework.
✅ RAG: Connects AI to your organization’s own documents and data so it generates answers grounded in what you actually know, not what the AI tool was trained on. Knowing this sharpens every vendor conversation about data access and integration.
In my AI adoption work, leaders who understand terms like these and “AI mechanics” write better policies, make sharper vendor decisions, and build/maintain effective governance routines.
⬆️ By reading this post, you’ve just leveled up your AI knowledge! Want to spread the word? DM me and I will send you a 1-page AI terms reference sheet you can share with your colleagues.
++++
A mid-size organization told me recently they had a solid AI governance policy.
Then we ran an AI audit. We found six AI features running across tools they already owned. One was summarizing HR conversations. One was scoring constituent records. None had been tiered for risk. None had a documented owner. Nobody had read the vendor data terms.
The organization’s AI policy was in place, but it only covered tools the team had deliberately chosen. It did not touch anything that arrived quietly in a software update.
Your AI policy is only as strong as the inventory behind it.
Three things to check this week:
✅ Open one platform your team uses daily
✅ Find the AI or “Intelligence” settings tab
✅ Ask: did we deliberately enable this, and who owns it if something goes wrong?
I run this inventory step with each of my clients before we build a governance framework.
DM me if you’d like a starter list of audit questions to address this common gap in governance.
++++
A global professional association Equifinality Solutions worked with recently had pretty good data.
Clean-ish. Current-ish. Stored across multiple systems where the same core metric was defined differently depending on who you asked.
When their data team learned the organization was pursing implementing AI-assisted querying of member data, their initial tests found accuracy to be inconsistent. AI was not wrong, exactly. It was working from multiple versions of the same truth.
Our engagement was structured around education first. We assessed their existing semantic assets against what AI-ready actually requires, then ran working sessions with their data team covering why ambiguous terms are the number one driver of natural language query failures, why one poorly defined table can cause an AI to count transactions instead of people and never flag the error, and how join logic needs to be encoded before an AI can use it reliably.
On the build side, we developed a business glossary that mapped ambiguous terms to single SQL definitions, a golden query set of validated question-to-answer pairs for their most common reporting questions, and a governance workflow with a clear RACI so the team could maintain what we built. We also documented their entity relationships and edge cases that would otherwise cause a language model to fail without warning.
The result: they are ready to reduce their data preparation time by 30% because they can focus on fortifying specific aspects of their data foundation.
This is the argument that contributors to a recent Amazon Web Services (AWS) Marketplace report on agentic AI make across 15 chapters and a dozen global enterprises. Tom Godden of AWS frames it directly in Chapter 1: the shift for AI is not volume times accuracy. It is context multiplied by connections multiplied by adaptability.
Chapter 11 makes the semantic layer concrete. Without a shared business vocabulary, even a well-configured LLM cannot interpret your context. It sees column names and schema labels, not the logic your organization actually runs on.
The organizations getting real value from AI right now built the foundation first. Governance, upskilling, and a clear adoption roadmap are key. If your organization is sitting on pretty good data that AI still cannot use effectively, DM me and I will share what Equifinality’s diagnostic typically surfaces.
++++
As an organizational psychologist focused on behavior change and AI adoption, I often see teams unsure of where to start using AI. Ethan Mollick‘s excellent guide, Using AI Right Now, offers a pragmatic starting point. I want to highlight three key actions (inspired by his guide) every team can take today:
🚀 Choose one tool and use it a lot
Try not to spread your attention across five tools. Pick one system (e.g., ChatGPT, Claude), subscribe to its most powerful model, and learn what it is capable of through experimenting with it. Talk openly within your team about what’s worked and what hasn’t. Aim to make AI fluency a shared team capability, not just the expertise of a few early adopters.
🚀 Use AI as a thinking partner
Instead of only using AI to draft content or summarize notes, use it to critique your work, challenge assumptions, or role-play stakeholders. For example, during a team meeting, invite team members to ask AI to develop alternatives for a decision you need to make, or adopt the persona of your stakeholder and offer best practices for a given activity. Then your team can discuss what pieces of AI’s input fit your situation the best, and build on those ideas together.
🚀 Lean into multimodal + research features
Use voice input: Walk and talk through challenges, like you would with a colleague. Ask the AI for input during a team meeting or individually while you are commuting to work.
Analyze documents or topics: Drop in reports, policies, or PDFs and ask the AI to extract insights, compare versions, or create summaries.
Perform on-demand research: Use AI to find up-to-date benchmarks, case studies, competitor activity, or funding trends.
Perform image-based analysis and/or reasoning: Upload charts, process maps, or whiteboard photos and ask the AI to interpret, clean up, or even redesign them.
Access the full guide here:
https://www.oneusefulthing.org/p/using-ai-right-now-a-quick-guide
🚀 Try one of these ideas and talk about what happens with your teammates. The future of work rewards experimentation!
++++
How can organizations get the most out of AI? The answer: reengineer the human systems around it. Workflows. Job roles. Team composition. Performance supports. Rewards. Metrics.
As discussed by Fortune‘s Jeremy Kahn, Microsoft’s research on companies they call “Frontier Firms” shows that AI success isn’t just about widespread use and adoption, it’s about redesign. These organizations prioritize impact over activity, restructure teams around outcomes, and rethink the very definition of value.
As an organizational psychologist and behavioral change expert, I help companies move beyond automation into true transformation. That means:
🚀 Flattening structures to reduce friction and increase quality of decision-making
🚀 Building intentional feedback loops
🚀 Creating space for AI to act as a partner (not just a tool)
🚀 Designing behavioral systems that guide people toward better focus, more fulfilling work, and strategic contribution
AI gives us the technology. Behavioral engineering is what turns that technology into results.
If we want AI to positively impact both individuals’ daily work AND organizational results, we must stop automating broken systems and start designing better ones.
++++
🚀 McKinsey & Company’s latest report highlights a growing disconnect in AI adoption: widespread use but minimal impact.
As an organizational psychologist and AI adoption specialist, I’m not surprised. The so-called “GenAI paradox” (where nearly 80% of companies are using generative AI but report no material gains) reflects a common behavioral pattern: layering new tools onto old workflows without rethinking how work should get done.
McKinsey rightly points out that real value lies in redesigning processes, not just automating existing tasks. But this shift isn’t just technical, it’s organizational and behavioral. It requires organizations to invite teams closest to the work to co-design new ways of working, rather than having an AI tool offered to staff and expecting increased productivity.
In my work across sectors, I’ve seen that success depends less on access to AI and more on the organization’s true readiness to redesign how work is performed. Companies that treat AI as an efficiency tool may find themselves outpaced by those that treat it as a catalyst for helping teams reinvent how they deliver results.
🚀 The message is clear: You don’t need more tools. You need a new mindset. Experimentation is the organizational skill of the future, and we must help teams reinvent how they work.
Access the report here:
https://www.mckinsey.com/capabilities/quantumblack/our-insights/seizing-the-agentic-ai-advantage
++++
🚨 Outdated workflows and team/job designs are holding back your ability to achieve gains from AI!
Many organizations are investing in AI but missing the productivity gains—because they’re trying to fit cutting-edge tech into legacy job structures. As MIT Sloan Management Review puts it: “Organizations must move beyond a simple binary narrative of substituting technology for the work that is currently being done by employees and embrace a fundamental redesign of work.” We need to stop thinking in terms of fixed roles and start redesigning work as a fluid system of human and machine tasks.
🚀 The path forward? Deconstruct tasks and workflows. Redesign roles and teams. Involve those closest to the work to ensure inclusive work practices and maximize talent. Reconstruct work from the ground up.
🚀 It’s time to work backward from the work, not forward from the tech.
Access the article here:
https://sloanreview.mit.edu/article/want-ai-driven-productivity-redesign-work/
++++
Selected Publications and Presentations
Justice Kelly, S. & Paul, M. (2026). AI for Nonprofit Leadership: Smarter Strategy, Stronger People. LinkedIn Blog Series.
Paul, M. & Smith, M. (2023). Handling the human factor: 3 considerations for building federal workforce technology skills. Reston, VA: ICF. https://www.icf.com/insights/technology/handling-human-factor-building-federal-workforce-tech-skills
Heinen, B. & Heelan, M. (2020). Post COVID-19: Improving your distributable workforce through formal telework programs. Fairfax, VA: ICF. https://www.icf.com/insights/workforce/improve-distributable-workforce-through-formal-telework
Heelan, M. (2020). 4 Ways to Easily Adapt In-Person Instruction to Virtual Learning. Fairfax, VA: ICF. https://www.icf.com/insights/culture/adapt-in-person-virtual-learning-covid-19
Heelan, M., Cavanaugh, K., & Lambourne, K. (2020). Making Results Count: Influencing Organizations to Achieve Population Results Through the Casey Children and Family Fellowship. Baltimore, MD: The Annie E. Casey Foundation.
Verive, J. M. & Paul Heelan, M. (2006). Using telework to achieve organizational outcomes: The effects of program formality. Presented at the 20th annual conference of the Society of Industrial Organizational Psychologists, Dallas, TX.
Schneider, B., Smith, D. B., & Paul, M. C. (2002). Attraction-selection-attrition model of organizational functioning. In M. Erez, H. Thierry, & U. Kleinbeck (Eds.), Work motivation in the context of a globalizing economy. Mahwah, NJ: Erlbaum.
Paul, M. C. & Lawson, C. (2001). iCulture: Transforming Human Capital into Economic Value. Presented at the 16th annual conference of the Society for Industrial and Organizational Psychology, San Diego, CA.