The GNL Podcast, Q&A
Episode 41, Q&A: CGM Testing Standards, What Grace Is, and AI in Healthcare
John spent two weeks on other people’s podcasts. Diabetech, then Tomorrow’s Medicine. He came back with listener questions that had not been answered on GNL before: what the international CGM testing standards debate means for people with diabetes in the UK, what Grace is at the architecture level, and where AI is going in healthcare work over the next two years. This is the Q&A.
Ask Grace
Curious what the IFCC stress test actually measures, what the bounded model means for how Grace reasons, or what the evidence says about AI and clinical roles?
Listen
Available on Buzzsprout, Apple Podcasts, and Spotify. Host: Anjanee Kohli. Guest: John Pemberton.
Why this episode exists
Listener questions came in after John’s two external podcast appearances in May 2026. On Diabetech, Gary Shiner’s show hosted by Justin, the conversation had been about CGM international testing standards: the IFCC stress test protocol that exists and works, the committee about to decide whether to require it, and the gap between what CE marking guarantees and what sensors are actually used for in insulin dosing. On Tomorrow’s Medicine with Arseniy Arsentyev, the conversation had been about Grace: the bounded model, the via negativa approach, and the near-bankruptcy moment that almost ended the project before it launched.
Both appearances left threads open. This episode closes them, in Q&A format: Anj puts the questions, John answers. No external guest. Forty minutes.
In this episode
There is no external guest. Anjanee Kohli hosts and John Pemberton answers listener questions across three topics: the CGM international testing standard debate (the IFCC stress test, what CE marking does and does not guarantee, and what listeners can actually do), what Grace is at the design level (bounded model, via negativa, no memory by design, and why), and AI and the future of healthcare work (which roles will adapt and which will not).
The episode was prompted by questions that came in after John’s appearances on Diabetech and Tomorrow’s Medicine. Both appearances are linked in the key themes section below.
Episode chapters
- 00:00, Introduction: a first-ever Q&A from listener questions
- 01:30, CGM testing standards: why the international debate matters and what you can do
- 15:13, What Grace actually is: architecture, philosophy, and no memory by design
- 25:49, The near-bankruptcy moment and the pay-it-forward model
- 31:28, AI and the future: which roles will adapt and which will not
- 39:13, Close
Watch or listen
Key themes
CGM international testing standards: the IFCC debate
The international committee that designs the standard for CGM accuracy is about to make a decision affecting everyone who uses an insulin-dosing sensor outside the United States. The IFCC has developed a stress test protocol for CGMs: a measurement of sensor performance during rapid glucose changes, with real insulin doses in the circuit. It tests the conditions that actually matter when a sensor is used to make dosing decisions. It exists. It has been validated. The committee is now considering whether to make it mandatory or optional.
The car crash analogy is the one that lands. Two cars, both approved, both passed the 30mph crash test. One also passed at 70mph. If your children were in that car, you know which one you would choose. The question is whether the people designing the standard think about it that way before the vote. The arguments for optional are that it is expensive and inconvenient for testing centres and requires manufacturers to run longer studies. The argument for mandatory is that sensors are being used to dose insulin during exactly the conditions the stress test was designed to measure.
John covered this on Diabetech, Gary Shiner’s show hosted by Justin, ahead of this episode. Watch that conversation: Diabetech episode on YouTube.
What Grace is: bounded model, via negativa, no memory
Grace is not a general-knowledge AI applied to diabetes. It is a bounded model: a defined evidence base, a defined scope, and a defined reasoning approach. The via negativa principle shapes how it reasons, working through what can be confidently excluded before stating what might apply. That is a different architecture from a tool that retrieves the most probable answer; it is one that removes the least defensible answers first.
The no-memory design is the one that still surprises people. Grace does not retain anything between conversations. That is not a product limitation: a tool with memory is one step closer to personalised recommendation, and the line between population-average educational guidance and personalised clinical advice is where Grace is designed to stay. Grace gets 80 per cent of the way there; the final 20 per cent requires the person with T1D and their diabetes care team.
John explained the Grace architecture on Tomorrow’s Medicine with Arseniy Arsentyev ahead of this episode. Watch that conversation: Tomorrow’s Medicine episode on YouTube.
Grace in 2026: three tiers, prices locked
Grace is free, and will stay free, for anyone living with T1D and their carers. The model that funds that promise has three inbound routes: a Grace HCP licence at £25 per month per licence for clinicians, ICBs, and manufacturer teams; a Grace Max licence at £60 per month per licence for research groups, manuscript leads, and audit teams; and a plus-one donation at £5 per month per slot via BuyMeACoffee for anyone who wants to sponsor a free account for someone with T1D. Every paid input is matched pound for pound by GNL at the platform level. Each input doubles. The free promise is structural, not aspirational.
AI and the future of work in healthcare
The third listener question was about AI and jobs. The framework John uses references Harari’s Nexus: the roles that will change soonest are the SOP-driven ones, where following a defined protocol is the primary task. Data processing, documentation, protocol-following. The roles that cannot be automated are the ones that require reading what is unsaid in a room: communication, clinical judgement from context, and the ability to be genuinely present with another person.
For healthcare professionals, the question is not whether change is coming. It is whether you will be in the 20 per cent who build and direct these systems, or the 80 per cent who adapts. The investment should go toward the skills that AI cannot reach.
No memory, by design. Grace does not remember previous conversations. That is not a limitation to be solved later. A tool that remembers your patterns and adjusts its answers accordingly is one step from giving advice that is personalised rather than educational. The absence of memory is part of how that line is held.
The CGM Series: context for the testing standards debate
The IFCC stress test debate connects to the GNL CGM Series. Episodes 35 and 36 built the accuracy-evaluation framework. Episodes 37, 39, and 40 applied it to three devices. The argument running through the series is that a consistent accuracy standard matters regardless of brand. The IFCC committee debate is the same argument at the international policy level: who decides what counts as a good enough test, and why it matters for the people using these sensors to dose insulin.
Episode 39, Abbott FreeStyle Libre 2 Plus and 3 Plus
Michael Skarlatos from Abbott. The 70 per cent size reduction and what it means for daily wear and stigma. The LibreView data nuance: a treated hypo does not appear in the low-glucose event log if glucose recovers before the 15-minute continuous threshold. The all-in-one applicator. Continuous ketone monitoring in development: simultaneous glucose and ketone measurement on the same sensor, alert-based rather than a continuous trace. Listen to episode 39.
Episode 40, Accu-Chek SmartGuide with Amy Jolley, DTN UK
17 people on the SmartGuide in one week, group onboarding, asked to report back. Night Low Predict: RAG system, purposeful rather than panicked responses, first-half versus second-half-of-night distinction. Glucose Predict’s 45-minute look-ahead versus the 15-minute trend arrow. The rage bolus named honestly. The DTN quality framework: five questions any CGM should answer, including 90 per cent in the 20/20 band and less than one per cent outside the 40/40 band. Listen to episode 40.
Grace in 2026: the three tiers
The Q1 quarterly review (episode 38) introduced Grace. Since then the model has been refined and the prices are locked. Three inbound routes fund the free promise; every paid input is matched pound for pound by GNL at the platform level.
What free Grace includes
Grace is free for anyone living with T1D and their carers. The free account includes 200 questions per calendar month (refills on the first), all six explorers, six modes, up to 2,000 tokens per response, one PDF per session, and 30 days of conversation history. No upsell, no countdown, no “upgrade” prompt. When the monthly allowance is reached, Grace shows “Grace is taking a breather” and nothing else.
What Grace HCP includes
Grace HCP is for clinicians, ICBs, named clinical teams, manufacturer-sponsored teams, and individual clinicians who want to recommend Grace in clinic with confidence in what it does. £25 per month per licence. Each licence activates one HCP account and funds two free Grace accounts for people with T1D. Enquiries to john@theglucoseneverlies.com.
What Grace Max includes
Grace Max is for research groups, manuscript leads, audit teams, manufacturer research and development, and evidence-base contractors. £60 per month per licence. Opus 4.7 model, all 20 modes, 8,000 tokens per response, file uploads, extended thinking, and a per-query cost cap. Each licence funds two free Grace accounts for people with T1D. Enquiries to john@theglucoseneverlies.com.
The plus-one model
Anyone can sponsor a free Grace account for someone with T1D. £5 per month per slot, via BuyMeACoffee. GNL matches pound for pound at the platform level. £5 in; two free Grace accounts out. Family members, charities, clinicians, manufacturers, and foundations all use it. The mechanism is pool-level, not assignment-level; there is no visibility of who receives a given slot.
Practical exploration
For people living with type 1 diabetes and their families
Three things from this episode worth acting on.
- The IFCC stress test debate matters for people using insulin-dosing CGMs. If you want to influence the outcome, Diabetes UK and Breakthrough T1D are the organisations to contact. Their members are the people this decision affects most.
- Grace is free if you are living with T1D. The app is at app.theglucoseneverlies.com. Grace does not remember previous conversations; every session starts fresh, and that is deliberate.
- If someone you know with T1D would benefit from Grace, the plus-one model lets you fund a free account for them for £5 per month, matched pound for pound by GNL. Two accounts out of every slot.
For clinicians and educators
Two things worth taking from this episode.
- The IFCC stress test debate is a patient-safety question in clinical language. If you work with people who use CGMs for insulin dosing, the gap between what the current CE standard tests and what sensors are used for in practice is worth knowing. Diabetes UK and Breakthrough T1D are the route if you want to influence the outcome.
- The Grace HCP licence is £25 per month per licence. Each licence activates one HCP account and funds two free Grace accounts for people with T1D. Enquiries to john@theglucoseneverlies.com.
About this episode
Anjanee Kohli is GNL’s Co-Director and Creative Lead, and a Specialist Diabetes Dietitian. John Pemberton is a paediatric Diabetes Dietitian at Birmingham Women’s and Children’s NHS Foundation Trust, GNL’s founder and director, and lives with T1D himself. This is a host-guest format with no external speaker: Anj hosts, John answers.
Related reading on GNL
Episode 41 of the GNL Podcast
CGM Series Review and Grace Update
This content is for educational exploration only. It describes average responses and general principles. It is not medical advice and cannot replace individual clinical guidance from your diabetes care team.