Why this episode exists

The last quarterly review was in December 2025. Since then, GNL has gone through more change than in the previous two years combined. The website has been rebuilt from scratch. Six interactive explorers are live on a Laravel API. GNL Grace, a bounded educational AI advisor with a curated evidence base, has launched. And a real-world validation study (33 structured assessments against approximately 500,000 patient-days from approximately 1,300 individuals, drawn from an upstream pool of over 10,000 individuals and 1.5 million patient-days collected 2013 to 2025) has been written up for publication.

This episode is John and Anj sitting down to make sense of it all; what happened, what surprised them, what went wrong, and what comes next. It is the first time John talks publicly about how Grace is funded; education paid for by industry grants and partner sponsorship, delivered free to every registered user.

A note on timing: this episode was recorded in late April 2026. The commercial structure described in the conversation evolved over the following weeks into the locked operating model now in place across GNL. The show-notes below reflect what Grace offers today; the audio captures the thinking that got us here.

In this episode

John and Anj recorded this as a fast, unscripted catch-up between ATTD in March and the GNL Grace launch. The Abbott FreeStyle Libre episode (originally scheduled as Episode 38) was held back while Abbott finalised details at their end, so the quarterly review stepped in. The conversation covers every major development from Q1 2026 and gives the first public look at how Grace works and why GNL exists.

What this episode explores

• The CGM guide update and CGM Series recap: Episodes 35 to 37, the five-out-of-five accuracy chart, and why only Dexcom G7, FreeStyle Libre 3, Roche Smartguide, and Medtronic Simplera currently qualify
• Two new sensors on the horizon (ICAN and CareSens) and what “seeing is believing” means for the accuracy threshold
• The new CGM Selector tool: eight attributes, pick your top three, get a recommendation
• How Phillip Hayes joined as Technical Director and built the explorer infrastructure
• The six GNL Explorers: what they do, how clinicians and people with diabetes are using them in practice, and why they get you 80% of the way there
• GNL Grace: a bounded educational AI advisor with a curated evidence base, not an unbounded chatbot pulling from the internet
• How Grace is offered today: free, always, for anyone with T1D and the people supporting them; Grace HCP licences from £25/month per licence for clinical workflow; Grace Max licences from £60/month per licence for research, deep evidence work and manuscript drafting
• How GNL is funded: three sponsor routes that mix any combination of plus-ones (£5/month per slot via Buy Me a Coffee), Grace HCP licences, and Grace Max licences. Every paid licence funds two free Grace accounts for people with T1D, one from the licence price and one matched pound for pound by GNL
• Real-world validation: 33 structured assessments against approximately 500,000 patient-days from approximately 1,300 individuals (T1D-adjudicated continuous-data subset of the upstream Syntactiq pool), and the publication that followed
• Why the skill of the future in healthcare is compassion, not knowledge, and what that means for how AI tools should sit alongside clinical consultations
• GNL merch: hats for the best feedback, and a Diabetes UK giveaway

Watch or listen

Episode chapters

• 00:00 – Introduction and the last few months in summary
• 01:34 – ATTD in March and the website redesign
• 01:55 – GNL Grace: the headline announcement
• 02:32 – CGM guide update: the five-out-of-five accuracy chart and new sensors on the horizon
• 03:53 – Abbott FreeStyle Libre episode held back; quarterly review steps in
• 04:48 – The CGM Selector: eight attributes, three picks, one recommendation
• 05:30 – Phillip Hayes: Technical Director, best mate, and the person who built the explorers
• 06:57 – How the explorers work: pop in your numbers, get the evidence-based answer for someone like you
• 08:10 – The 80/20 split: explorers handle the knowledge, clinicians handle the person
• 09:40 – GNL Grace: the six-layer bounded AI advisor, not a chatbot
• 10:50 – How Grace is offered: free to every registered user, with audience self-identification at sign-up
• 12:34 – What a Grace session looks like: full chat interface with the explorers side by side
• 13:31 – Grace Max: the higher-power tier, now publicly priced at £60/month per licence (this episode was recorded under the earlier invitation-only framing, retired 10 May 2026)
• 17:31 – Real-world validation: approximately 500,000 patient-days from approximately 1,300 individuals across 33 structured assessments, and the upcoming publication
• 19:58 – The skill of the future is compassion, not knowledge
• 22:27 – How explorers and Grace change the clinic consultation
• 24:51 – What surprised John most: the depth of analysis Grace can produce
• 30:09 – What worked, what broke, and why login emails went out three times
• 33:57 – Feedback and evolving Grace: adding modes, expanding concepts
• 35:06 – GNL merch: hats for the best feedback
• 36:10 – Closing: what comes next, and Phil’s episode

Key themes

1. The website rebuild: from 150 pages to something people can actually navigate

GNL had 150 pages of content that John describes as genuinely hard to find. People were arriving on the site with a question and struggling to locate the answer. Phil’s contribution was not to add more content but to change how people interact with it. The explorers, the CGM Selector, and ultimately Grace are all designed around the same principle: arrive with a question, leave with an answer, in as few clicks as possible. The site went through what Anj describes as “about six facelifts, probably in the last two weeks” and the architecture behind it is now built on a Laravel API rather than static pages.

2. The GNL Explorers: six tools, one principle

The six explorers (Activity, Exercise IOB, Hypo Treatment, Hyper Treatment, AID Algorithm Optimiser, and Exercise Planning) all work the same way. You enter a few personal details; your weight, total daily insulin dose, and a couple of preferences. The algorithm returns what the evidence says for the average person with those parameters. John is emphatic that this gets you 80% of the way there. The other 20% is the clinician’s relationship with the individual; their preferences, their circumstances, the things an algorithm cannot know. The explorers have been tested in clinic and the feedback from clinicians is that they save time on the number-crunching, freeing up the consultation for the conversation that actually matters.

3. GNL Grace: a bounded system, not a chatbot

John draws a clear line between Grace and general-purpose AI tools like ChatGPT. Grace is bounded: the evidence base is curated by John, built in layers (ISPAD and ADA guidance at the base, GNL guides and framing above that, deterministic explorer logic, then curated research on top), and nothing from the open internet gets in. When a new paper is considered, the best 100 are narrowed to 10, then only five go into the wiki. The result is an advisor whose answers come from a controlled, high-quality evidence base rather than from whatever the internet happens to contain. Grace currently sits on around 70 concept trees and the structure is designed to scale.

4. How Grace is offered, and how GNL is funded

Grace is free to every registered user. Sign-up is email-only (no card), and the user self-identifies as part of one of two audiences: people with diabetes and their carers, or healthcare professionals, researchers, and industry partners. Both audiences get the same Grace, with framing tailored to context.

Grace HCP is the licence tier built for clinical workflow, learning, teaching, and reference. From £25/month per licence, with generous monthly limits, all six Explorers, and full clinical-context routing. Bought by Trusts, ICBs, named clinical teams, manufacturer-sponsored teams, and individual clinicians.

Grace Max is the higher-power tier for research, deep evidence work, manuscripts, comparing systems, and quality-assuring content. From £60/month per licence, with all twenty modes, file uploads, extended thinking, and the Spider-Man powers framing. Bought by research groups, manuscript leads, audit teams, manufacturer R&D, and evidence-base contractors. The earlier “by invitation only” framing was retired on 10 May 2026; Max is now publicly priced.

The funding model runs on three sponsor routes that anyone can mix. Plus-ones at £5/month per slot via Buy Me a Coffee (any donor, any volume; GNL matches each slot pound for pound). Grace HCP licences from £25/month per licence. Grace Max licences from £60/month per licence. Every paid licence funds two free Grace accounts for people with T1D, one from the licence price and one matched by GNL. No public price sheet; volume discounts negotiated in conversation. The earlier “three commercial routes” framing (educational grants, sponsorship of explorers, Grace Max access) was consolidated into this unified sponsor-route model on 10 May 2026.

5. Real-world validation on approximately 500,000 patient-days from approximately 1,300 individuals

A founder in the diabetes data space, having seen the GNL alcohol and drugs guide, offered John 24 hours of access to a longitudinal dataset (T1D-adjudicated continuous-data subset of approximately 1,300 individuals representing approximately 500,000 patient-days, drawn from an upstream Syntactiq Dynamics FlexCo pool of over 10,000 individuals and 1.5 million patient-days collected 2013 to 2025). John ran 33 structured assessments of the GNL algorithms against that data. Most passed; the handful that did not led to algorithm adjustments. The paper has been written up with diagrams generated through the Grace Max workflow and is heading for submission. This validation is what gave the educational positioning confidence that Grace operates as an evidence-based educational tool, not a clinical device making unsupported claims.

6. The 80/20 principle: AI handles knowledge, humans handle people

This is the thread that runs through the entire episode. John’s position is specific: AI will automate knowledge retrieval and evidence synthesis faster and more consistently than any human. That is real, valuable work. But it is not the thing that changes outcomes for individuals. The skill that cannot be automated is the ability to sit with a person, understand their circumstances, and help them make a decision that fits their life. If Grace handles the 80% (the evidence, the algorithm matching, the “what does the research say”), then the 20% that remains is the part that actually changes outcomes; and it gets the attention it deserves instead of being squeezed into a 10-minute appointment.

7. The skill of the future is compassion, not knowledge

John makes this case directly to healthcare professionals who might feel threatened by AI tools. His argument is not that AI is harmless, or that professionals have nothing to adapt to. It is that the skill being automated is knowledge retrieval, and the skill that cannot be automated is compassion, understanding, and delivery. If you are starting a career in diabetes education or clinical practice, the smartest investment is not in memorising every algorithm. It is in becoming the person who can sit with someone and help them figure out what matters to them. The “knowledge robot healthcare professionals” will eventually be outpaced; the ones who are genuinely good with people will not.

8. ATTD and the conversations that changed the direction

John describes ATTD in March 2026 as the inflection point. Conversations at the conference, particularly around digital twins, changed how he thought about what was possible with educational technology. The return from ATTD coincided with Phil introducing Claude Code, and the two things together triggered an intensity of development that John did not predict in the December quarterly review. He is honest about that: the December episode laid out plans, but the pace at which those plans evolved into what GNL is now was not expected.

9. What went wrong: registration emails and the cost of moving fast

John is candid about the registration launch. The password reset did not work. Subscribers got emails, tried to log in, and were locked out. The fix was deployed, the problem recurred, and the fix was deployed again. John’s position is that low-risk, non-clinical things (like a login flow) are worth pushing out fast and fixing in public, because the alternative is spending days on verification that could be spent building. The trade-off is that a few people get annoyed. He acknowledges that, apologises for it, and makes the case that GNL Grace does not happen without that pace.

10. GNL merch and the Diabetes UK giveaway

GNL has branded hats (black and white) and the plan is simple: give the best feedback on the explorers or on Grace, get a hat. Ali Balderstone and Fredrik (the founder who provided the dataset access) were the first two. For anyone heading to Diabetes UK the following week, John’s offer was direct: use Grace, screenshot the best answers, email them over, and come find him at the conference for a hat. It is a small thing, but it is the kind of feedback loop that makes the product better.

Practical exploration checklist

For people with type 1 diabetes (and their families):

• If you have not visited the GNL website recently, start with the GNL Explorers; pick the one closest to your question and enter your details for an evidence-based summary at population level
• Try the CGM Selector if you are choosing between sensors; pick your three most important attributes and see which CGMs match
• Create a free GNL account at app.theglucoseneverlies.com to access Grace; chat interface, evidence-graded answers, with framing for people with diabetes and their carers
• If Grace gives you an answer that does not feel right, trust your instinct; she gets you 80% of the way at population level, but she does not know you personally
• Bring interesting Grace outputs to your next clinic appointment as a starting point for discussion with your care team

For clinicians, researchers, and industry:

• Explore the AID Algorithm Optimiser with someone you support in mind; see what the evidence-based neutral settings look like for their system, then adjust from there with your team’s clinical judgement
• Consider how the 80/20 model could change your consultations: if the explorer handles the algorithm lookup, what would you do with the extra time?
• Register at app.theglucoseneverlies.com as a healthcare professional and try Grace with a clinical question you would normally spend time researching; compare the answer quality and the evidence grading to your usual process
• If your organisation wants to sponsor Grace (HCP licences, Max licences, or plus-ones for the T1D community), email john@theglucoseneverlies.com

About the hosts

John Pemberton is the founder and director of The Glucose Never Lies, a paediatric Diabetes Dietitian at Birmingham Women’s and Children’s NHS Foundation Trust supporting over 300 children with T1D and their families, a published researcher, and a structured-education clinician across DAFNE, BERTIE, DYNAMIC, and GAME programmes. He built the GNL evidence base, curates every citation that enters Grace’s wiki, and developed the algorithms behind the explorers. He lives with type 1 diabetes himself; the lived lens and the family lens (Jude tested antibody-positive on the Type 1 newborn spot screen and negative on the follow-up blood draw, the founding moment that started GNL) both run through everything GNL writes.

Anjanee Kohli is Co-Director Creative at GNL. A registered dietitian with a background in diabetes education, Anj leads the podcast production, social media, newsletter, and visual identity across all GNL platforms. She is the reason episodes sound professional, look consistent, and go out on time.

Related reading on GNL

• GNL Grace – the bounded educational AI advisor discussed throughout this episode, with the methodology and evidence layers explained
• GNL Explorers – all six interactive tools in one place
• AID Algorithm Optimiser – the explorer John describes using in clinic
• The GNL CGM Series – Episodes 35 to 37, with the Abbott episode coming next
• Episode 37 – Dexcom G7 and One Plus – the most recent CGM Series episode
• The GNL Podcast Hub – every episode in one place