After months of writing, stats, caffeine, and existential dread... my dissertation has officially landed on ProQuest! I studied how AI code generation affects developer productivity so you don’t have to. proquest.com/docview/3217383… #DeveloperProductivity #AI

Why Human-in-the-Loop Isn’t Optional There’s a common idea in AI: “The goal is full automation.” In practice, that’s rarely how successful systems work. Most real AI systems rely on: 👉 human-in-the-loop design Not as a fallback. As a core part of the system. Here’s why: 1. Trust doesn’t exist by default Users need to see, validate, and understand outputs before relying on them 2. Edge cases are everywhere No model handles every scenario cleanly 3. Context still matters Humans bring situational awareness that systems don’t fully capture 4. It enables learning Human feedback improves the system over time Where companies go wrong: They design AI as: Fully automated from day one Or completely manual with no leverage The reality is in between. Good systems: Use AI to accelerate decisions Keep humans where judgment matters Gradually increase automation over time Human-in-the-loop isn’t a weakness. It’s how systems actually scale. If you’re building or deploying AI right now, this is one of the most important design decisions you’ll make. I’ll go deeper into this in future posts. If you want a structured way to approach system design, I’ve outlined it here: 👉 drscottmorgan.com

Decision Support vs Automation Most companies are trying to use AI to automate everything. That’s the wrong starting point. Not every decision should be automated. The real question is: 👉 Should this be decision support… or full automation? Here’s how I think about it: Decision Support High cost of being wrong Requires human judgment AI provides recommendations, not final decisions Automation High frequency Low variability Low cost of failure (or well-controlled) Clear inputs and outputs Where companies go wrong: They try to automate decisions that: Are ambiguous Carry high risk Require context the model doesn’t have And when it fails, trust is lost immediately. A better approach: Start with decision support Build confidence Then selectively automate AI isn’t about replacing humans everywhere. It’s about placing the system at the right point in the decision flow. That’s where the real leverage is. If you’re working through this tradeoff, I’d be interested to hear how you’re approaching it. And if you want a structured way to think about these decisions, I’ve outlined my approach here: 👉 drscottmorgan.com

Why Most AI Use Cases Are Wrong Most AI use cases sound good. That’s the problem. They’re framed as: “Let’s use AI to improve X” “Let’s build a chatbot for Y” “Let’s automate this process” But they’re missing one thing: A decision. AI is not about features. It’s about decisions. If your use case doesn’t clearly answer: 👉 What decision is being made differently? …it’s probably wrong. Strong AI systems: Support or automate decisions Reduce cycle time Improve consistency Change outcomes Weak ones: Add another tool Sit outside the workflow Depend on user choice Don’t impact real metrics This is why so many AI initiatives stall. They’re not tied to how the business actually runs. They’re layered on top of it. The shift is simple, but critical: From: “Where can we use AI?” To: “Where are decisions being made at scale?” That’s where the real leverage is. That’s where AI works. I’ll break this down further in future posts. If this is something you’re thinking through, I’ve put a structured approach together here: 👉 drscottmorgan.com

Most AI projects don’t fail; they stall. This is often more detrimental because a stalled AI initiative can appear to be making progress. There may be a pilot, the model may “work,” and there could be demos and updates, yet nothing changes in the business. There is no real adoption, no workflow integration, and no measurable impact. I've observed a recurring pattern with stalled initiatives: 1. The use case was never strong enough - It sounded interesting but lacked real operational leverage. 2. No one actually owned the outcome - There may be an “AI team,” but there is no accountable operator. 3. It lives outside the workflow - Users must choose to use it, and often they don’t. 4. It’s not tied to financial results - Consequently, it never becomes a priority. At this stage, the initiative doesn’t fail; it simply sits there. I refer to this as pilot purgatory, and many companies find themselves in it. Scaling AI isn’t just about improving the model; it’s about addressing: - Ownership - Integration - Measurement This distinction is crucial between a working demo and a functioning business system. I will break down how to avoid this in upcoming posts. If this resonates with you, you are not alone. For those looking to move beyond pilots into something tangible, I’ve outlined my approach here: 👉 drscottmorgan.com

AI doesn’t fail because the models don’t work; it fails because organizations don’t know how to scale it. Over the past year, I’ve been immersed in enterprise AI delivery, focusing on real systems, workflows, and financial outcomes. Here’s what I’ve observed: Most companies aren’t struggling with AI capability; they’re struggling with AI implementation. They build: - Impressive pilots - Smart models - Interesting demos …but they often fail to translate that into: - Workflow integration - Adoption - Measurable business impact As a result, initiatives stall—not because the AI failed, but because the supporting system never existed. After witnessing this pattern repeatedly, I started structuring the problem to connect: - Use-case selection - System design - Workflow integration - Governance - Financial outcomes I call it the AI Scale Framework (AISF). This framework helps move AI from experimentation to production and ultimately to EBITDA impact. In the coming months, I will break this down in detail, discussing what works, what doesn’t, and what actually scales in real organizations. If you’re navigating AI in your company, I’d like to hear where you’re facing challenges. For those interested in a deeper dive, I’ve compiled the full framework here: drscottmorgan.com

How easy is it to create a lesson plan in Plansora? Find out here - youtube.com/watch?v=ixwWppOg…

AI data preparation is transforming analytics workflows. By automating data cleansing and integration, organizations can save up to 80% of the time spent on these tasks, allowing data scientists to focus on higher-value initiatives. #AIinData

In exploring AI code generation, a recent study reveals that developer productivity can increase by up to 50% when incorporating intelligent tools. This shift highlights the evolving role of developers as they transition from writing to refining code. #devproductivity

Retrieval Augmented Generation (RAG) is enhancing the capabilities of language models by integrating external knowledge sources. This development promises more accurate and contextually relevant outputs, which is crucial for applications in enterprise. #RAG

Retrieval Augmented Generation is enhancing natural language processing by combining retrieval techniques with generative AI. This hybrid approach promises improved context and accuracy in generated outputs, reflecting a significant shift in AI capabilities. #RAG

devs joining standup at 8am

Every classroom is different. That’s why Plansora lets you choose your lesson plan format and instantly generate plans powered by TEKS standards. 📝✨ plansora.com Customizable. Aligned. Ready in minutes. #TeacherTwitter #EdTech #AIart

Teachers in Texas know the importance of TEKS. 📚 With Plansora, you can browse TEKS standards and build AI-powered lesson plans aligned to exactly what your students need. plansora.com Save time. Stay aligned. Teach with confidence. #EdTech #TEKS #AIinEducation

🚀 Excited to share what I’ve been building: Plansora. An AI-powered lesson plan generator designed to save teachers hours of prep time. Imagine creating standards-aligned lesson plans in minutes instead of hours. plansora.com Early preview 👇 #EdTech #AI

If ChatGPT can help write essays… …why not lesson plans? That’s exactly what I’m building with Plansora. AI-generated, TEKS-aligned plans to save teachers time. plansora.com Want early access? Follow along 👇 #EdTech

AI-driven Data Preparation is reshaping how organizations handle their datasets. These tools streamline the cleansing and transformation processes, ultimately enabling more accurate analytics and informed decision-making. A game changer for data-driven strategies. #DataPreparation

Teachers spend countless hours planning lessons. 📚 What if AI could cut that down to minutes? That’s the mission of Plansora — an AI tool that builds TEKS-aligned lesson plans fast. plansora.com More to come soon. Stay tuned! #EdTech #TeacherLife

Retrieval Augmented Generation is redefining how we approach data interaction. By enhancing the generative capabilities of AI with specific data retrieval techniques, we're witnessing a marked improvement in contextually relevant outputs. Innovation is key here. #AI #RAG