by Gary Mintchell | Apr 6, 2026 | Business, Generative AI
I receive great value from the wisdom and generosity of Seth Godin. He thought out this progression of business value. As you create products and companies and personal value, consider this deeply and seriously—Create value by connecting people.
From Seth Godin Feb 13
The first generation was built on large models, demonstrating what could be done and powering many tools.
The second generation is focused on reducing costs and saving time. Replacing workers or making them more efficient.
But you can’t shrink your way to greatness.
The third generation will be built on a simple premise, one that the internet has proven again and again:
Create value by connecting people.
We haven’t seen this yet, but once it gains traction, it’ll seem obvious and we’ll wonder how we missed it.
Create tools that work better when your peers and colleagues use them too. And tools that solve problems that people with resources are willing to pay for.
Problems are everywhere, yet we often ignore them.
Click on the Follow button at the bottom of the page to subscribe to a weekly email update of posts. Click on the mail icon to subscribe to additional email thoughts.
by Gary Mintchell | Mar 31, 2026 | Generative AI, Podcast
I’ve released a new podcast.
You can subscribe and download from your favorite podcast app or from my site.
It is also available on my YouTube channel.
Episode 275. Why AI in Manufacturing? Why not? I explore how new technologies for knowledge work, unlike in manufacturing, create even more busy work distracting us from our real work–thinking, deep work. Looking beyond the hype, AI tools are going to help us do things. We just don’t know exactly what is best. We must play with the tools to find the best ways to help us.
We also need to consider the limits of text-based LLMs. Researcher Yan LeCun has looked at the limits of these technologies. How can they work for things like bringing a robot into the house, for example, when they are limited to digital and text and the environment is analog. Won’t this take a system of models? Not just one model?
Use them, but don’t be awed. Or bamboozled by CEOs.
This episode is sponsored by Inductive Automation.
Click on the Follow button at the bottom of the page to subscribe to a weekly email update of posts. Click on the mail icon to subscribe to additional email thoughts.
by Gary Mintchell | Mar 31, 2026 | Generative AI, News, Security
Several people involved with standards have shared with me the insight that the driving force for adoption of some of these will come from company boards due to insurance and risk management pressures. Therefore, I found this paper interesting looking at trustworthy AI from the point-of-view of risk management.
Høvik, Norway, 25 March 2026 – New research from assurance and risk management company DNV has identified the foundations to achieving trustworthy artificial intelligence in the context of safety critical industrial processes. According to the paper, Assurance of AI-Enabled Systems, established risk management principles can be adapted to meet the complexity and uncertainty of AI enabled systems. While AI introduces new risks, proven assurance methods from safety critical industries already provide a robust starting point for addressing them
The paper shows that AI reshapes risk because it does not operate as a fixed, predictable component. This makes traditional one‑time assurance insufficient, and highlights the need for continuous and adaptive assurance throughout the lifecycle
Christian Agrell, Programme Director for AI Assurance at DNV, said, “Creating trustworthy artificial intelligence does not require us to start from zero. We already have strong foundations in modern assurance and risk science and our long experience managing digital technologies in high‑risk environments. Applying these principles thoughtfully allows us to build systems that remain safe and reliable, even as they evolve. Trustworthy AI depends on predictable behaviour under uncertainty, and that is exactly what these foundations help deliver.”
The research draws on DNV’s decades-long assurance and risk management experience in critical infrastructure, including the maritime and energy sectors. The foundational principals to create trustworthy AI include:
- A system model that captures the entire AI-enabled system
- This model reflects how AI interacts with humans, digital and physical components, and its operational environment. It enables understanding of emergent behaviour, unintended interactions and context specific risks that cannot be detected by examining the AI component alone.
- Taking a modular approach
- A risk model, applying uncertainty-based assessment and modular risk principles to break down complex systems with their complex and emergent risks into manageable parts across system levels.
- Linking claims to evidence
- These structured arguments connect claims such as “the system is safe” to verifiable evidence, assumptions and rationale. This provides a transparent, auditable framework for demonstrating trustworthiness throughout the lifecycle.
- Continuous, context aware assurance that adapts as AI evolves
- AI-enabled systems change over time as models are updated, data shifts and operating conditions vary. To maintain trustworthiness, assurance must be ongoing rather than a onetime check. This includes real-time monitoring, regular updates to evidence, and reevaluating risks and requirements so that confidence in the system remains valid throughout its lifecycle
“These foundations give industry a clear, actionable way to build and maintain trustworthy AI. We are already working with companies that recognize the potential of AI, as well as the risks it can pose to the critical services they deliver. I urge more organizations to join us in addressing and managing the risks associated with artificial intelligence,” Agrell added.
The position paper is part of DNV’s broader work to help industry adopt AI responsibly and aligns with the company’s recommended practice (DNV‑RP‑0671) for AI assurance.
Click on the Follow button at the bottom of the page to subscribe to a weekly email update of posts. Click on the mail icon to subscribe to additional email thoughts.
by Gary Mintchell | Mar 26, 2026 | Generative AI, Process Control, Software
TwinThread is one of those smallish software companies within an interesting niche that I can’t believe has yet to find a buyer. I quoted noted software developer and LinkedIn commentator Rick Bullota in 2020 extolling the value of the AVEVA/TwinThread link with the AVEVA purchase of OSIsoft. Just last year, I wrote about a stronger partnership between the two.
I see the company has pivoted a bit to now proclaiming itself as “the world’s first to have a complete Industrial AI platform.” I’ll leave that proclamation to your judgement. But this product looks worthwhile to check out.
Last week’s news involved TwinThread releasing an AI-powered manufacturing analytics solution targeting batch processes called Perfect Batch. This product empowers manufacturers to standardize and consistently replicate their best performing or “golden batches”.
Perfect Batch applies industrial AI to dynamically identify ideal batch profiles from historical data and actively recommend actions for increasing efficiency. This enables organizations to rapidly shift from reactive firefighting to proactive optimization in a matter of weeks – not months or years.
Perfect Batch At-a-Glance:
- Rapid Speed to Value: Perfect Batch connects to existing batch execution systems and automatically interrogates past data to build digital twins and apply models in hours.
- Dynamic Perfect Profile Learning: Instead of setting limits manually, Perfect Batch dynamically learns ideal control limits and process centerlines, based on actual process capability and historical performance.
- Unlocked Hidden Capacity: Granular cycle time analysis identifies bottlenecks and lost production time, facilitating capacity improvements from existing assets without new capital investment.
- Optimization by Exception: Automated alerting and issue diagnosis empowers operations teams to focus on solving problems, without getting bogged down with endless troubleshooting and investigations.
- Optional Closed-Loop Action: Thread Builder, a real-time workflow engine that works with Perfect Batch, automates anomaly responses, performs automatic diagnoses, and can trigger specific corrective actions automatically.
- Automated Compliance: Tailored for regulated industries, Perfect Batch provides automated material tracking, quality and yield conformance, and audit-ready histories.
Beyond the plant floor, Perfect Batch helps drive strategic and collaborative alignment across organizations’ entire manufacturing portfolios by providing a global view of asset utilization and batch making performance. As a result, the platform serves as a single source of truth for cross-functional teams. This offers a common lens that operations teams, engineering teams, and supply chain leaders can all use to identify, prioritize, and proactively execute improvement initiatives that optimize the deployment of capital across the supply network.
Click on the Follow button at the bottom of the page to subscribe to a weekly email update of posts. Click on the mail icon to subscribe to additional email thoughts.
by Gary Mintchell | Mar 19, 2026 | Generative AI, Software
Notes and news about AI continue to build in my pending folder. Too many to figure out. I’ll start with this one. I saw this news item from Morning Brew, one of my daily news feeds…The AI…it filled the code with bugs.
The amount of bugs popping up in AI-generated code is reaching the loose Sour Patch Kids under a camper’s bunk level. Amazon’s e-commerce senior VP, Dave Treadwell, called an all-hands for engineers at the company yesterday to address the growing frequency of outages, some of which can be traced back to code developed by generative AI, according to the Financial Times.
It continues…
- Last week, Amazon’s store malfunctioned for a few hours, which the company attributed to “a software code deployment.”
- And Amazon’s cloud services unit, AWS, had at least two large outages recently related to AI coding assistants. In December, the company’s cost calculator was down for 13 hours when Kiro, its AI coding tool, tried to change the code, and delete and remake the entire system.
- Though Amazon downplayed the meeting as routine in comments to the FT, the paper reported that Treadwell told employees that senior engineers will now need to sign off on AI-assisted changes made by junior and mid-level engineers.
Solutions?
An expensive solution. Anthropic rolled out a review tool yesterday in Claude Code to (hopefully) catch those vibe-coded mistakes—but with each pull request costing up to $25, it may get pricey fast.
Concurrently with this news, I received a PR request to interview Pramin Pradeep, CEO of BotGauge AI. I receive this sort of thing many times daily. Supposedly, Pradeep wanted to talk about “shadow code” left behind, supposedly maliciously, by AI generated code.
I asked for something in writing. They sent the usual PR thing that mentions shadow code but switches the topic to cybersecurity and then cites an irrelevant “case study”.
However, BotGaugeAI does participate in a market (Claude told me they were 20 out of 128 in that market for what it’s worth) called AI-assisted QA for code.
My research revealed that the basic problem comes from where the LLM AI code was trained. If trained too broadly, it will tend to being in superfluous code. Managers, meanwhile, are discovering that while maybe coders can save time in development using LLMs the task of checking and approving is becoming onerous.
If you’re using LLMs to help code, it would probably pay to check out the companies like BotGaugeAI for automated QA.
Click on the Follow button at the bottom of the page to subscribe to a weekly email update of posts. Click on the mail icon to subscribe to additional email thoughts.
by Gary Mintchell | Mar 18, 2026 | Generative AI, Robots
Now that we’re learning more about Large Language Model AI, users have discovered that what the model is trained on is one important key. That makes this robot AI trainer intriguing.
Universal Robots and Scale AI Launch Imitation Learning System to Accelerate AI Model Training, Bridging the ’Lab-to-Factory’ Gap
Universal Robots (UR) unveiled March 16 the UR AI Trainer at GTC 2026 in Silicon Valley. Developed in collaboration with Scale AI, the AI Trainer marks a tectonic shift as robots move from pre-programmed applications to fully AI-driven tasks. These systems are powered by robust data generated in AI training cells where robots imitate humans.
“Our customers, ranging from large enterprises to AI research labs, are no longer just asking for AI features,” said Anders Beck, VP of AI Robotics Products at Universal Robots. “They need a way to collect high-fidelity, synchronized robot and vision data to train AI models on the same robots they intend to deploy. Our AI Trainer is the industry’s first direct lab-to-factory solution for AI model training.”
Alongside the new AI Trainer, Universal Robots’ GTC booth will showcase a state-of-the-art robotic foundation model from Generalist AI, a UR preferred model partner. Leveraging this model, two UR robots will complete a complex smartphone packaging task, previously impossible without recent advances in the field of Physical AI.
AI robotics training is often hindered by fragmented hardware and low-fidelity data capture. Much of today’s training data is collected on research robots not suited for production environments, and many systems rely only on visual feedback, making delicate or contact-rich tasks difficult. “The AI Trainer directly addresses these barriers,” said Beck. “By utilizing our unique Direct Torque Control and force feedback features, we give developers direct influence over how the robot physically interacts with the world, training on the same robust hardware used in over 100,000 industrial deployments.”
The AI Trainer allows human operators to guide UR robots through tasks in a leader-follower setup while automatically capturing high-quality multimodal data for robotics AI development. Operators physically guide a “leader” robot through a task while a synchronized “follower” robot mirrors the motion in real time. During each demonstration, the system records synchronized motion, force, and visual data, producing the structured datasets required to train Vision-Language-Action (VLA).
Deploying on UR’s AI Accelerator platform, the UR AI Trainer combines UR robots with Scale AI software to enable data capture on UR robots in production and at scale creating continuous feedback that drives ongoing optimization of physical AI systems.
With GTC as the official launch pad, attendees will be able to experience the system first-hand at UR’s booth as they guide two UR3e ‘leader’ robots providing haptic input to control two UR7e ’follower’ robots. The setup enables visitors to perform an advanced smartphone packaging task with haptic feedback for imitation learning and VLA training, with demonstration data recorded in real time on Scale’s stack and replayable directly on the AI Trainer.
The process of capturing robot training data for AI models is further showcased through a demo that illustrates the same smartphone packaging task – just trained virtually: Built in NVIDIA Omniverse and leveraging Isaac Sim, the simulated setup allows attendees to control a virtual bi-manual UR3e system with real-time haptic feedback using two Haply Inverse3 devices as ‘leaders’, providing a physics-accurate simulation.
Universal Robots is also exploring the use of the NVIDIA Physical AI Data Factory Blueprint to automate and scale its synthetic data generation, transforming world-scale compute into a production engine for high-quality robotic training data.
Complementing the two data-capture demonstrations, Generalist’s showcase highlights how advances in data collection and AI models translate into real-world robotic performance. In the first public demonstration of Generalist’s embodied foundation models, two UR7e robots autonomously execute a complex smartphone packaging task, demonstrating dexterity, coordination, and contact-rich manipulation in a real-world environment. The demonstration shows how scaled, high-quality training data combined with frontier model architectures can enable robust physical AI systems beyond the lab.
Click on the Follow button at the bottom of the page to subscribe to a weekly email update of posts. Click on the mail icon to subscribe to additional email thoughts.
