Tom Loosemore
@tomski.com
Partner at Public Digital, co-founder Government Digital Service, ex-BBC. Expect occasional small boat sailing.
Except reliability, which is improving, but v slowly hal.cs.princeton.edu/reliability/
HAL Reliability Evaluation
Accuracy isn't enough. We evaluate AI agents on reliability — consistency, predictability, robustness, safety, and self-awareness.
hal.cs.princeton.edu
One of my smartest friends pointed out that an important assumption in this post was already wrong. So I've added an update at the top.
Tom Loosemore @tomski.com · 2d agoI've written a long blog post about AI Agents and the need for those running services to prepare - technically, operationally, commercially, strategically, legally and in terms of policy. AI Agents could be a properly huge deal. There. I wrote it. loosemore.com/2026/08/04/k...
Done right, AI really can help you write good software faster. But writing software faster should never get into a top 10 list of factors enabling the creation of successful new services at pace.
I've written a long blog post about AI Agents and the need for those running services to prepare - technically, operationally, commercially, strategically, legally and in terms of policy. AI Agents could be a properly huge deal. There. I wrote it. loosemore.com/2026/08/04/k...
Know Your Agent: Why you should act now, to avoid playing catch-up later
Back in March, I was beaten by the bins. I’d spent a happy few evenings (and a great many tokens) trying to build a simple web app: share any UK address, find out which bins to put out, and w…
loosemore.com
Know Your Agent is gonna be a big deal. Messy, but important. loosemore.com/2026/07/29/k...
KYC, KYB and their wild new cousin, KYA
Earlier this year I blogged about the idea for a Creative Commons-style legal framework to help services respond to requests from AI Agents claiming to act on behalf of a person. The more I play wi…
loosemore.com
Termsync.co.uk is most excellent, not least it's running on an open weight LLM hosted under the maker's desk.
TermSync — School term dates, nailed
Subscribe to your child's school term dates. Terms, half-terms, INSET days and holidays land in your calendar and stay up to date on their own.
termsync.co.uk
ICYMI a long read on how digital is missing from Burnham's vision for how to govern. I would be very comfortable if his first flip flop came by revisiting the decision to send digital to the margins. It's a bad call and it seems to be the consensus view among people who know about this stuff.
Benjamin Welby@bm.wel.by · 2w agoBy popular* demand I have added my many words to all the other bits and pieces reflecting on the end of DSIT. *@semanthicc.bsky.social and one other said they were waiting to hear my take.
Simon Willison is an (inter)national treasure. If you want the definitive take on the OpenAI/HuggingFace hack, it's here on his excellent blog: simonwillison.net/2026/Jul/22/...
OpenAI’s accidental cyberattack against Hugging Face is science fiction that happened
This story is wild. The short version: OpenAI were running a cybersecurity test against an unreleased model, with the model’s guardrail features turned off. Rather than solve the test, the …
simonwillison.net
Looking forward to some rain, not least it'll be me an excuse to play with my new toy.
By popular* demand I have added my many words to all the other bits and pieces reflecting on the end of DSIT. *@semanthicc.bsky.social and one other said they were waiting to hear my take.
GDS, AI and the missing theory of the digital state
Two years after digital, data and AI were brought together, the settlement is being dismantled. What does that tell us about the state Britain is trying to build?
bm.wel.by
Incubator for AI → DCMS AI adoption and preparedness + Frontier and Emerging Technologies including NSSIF and SovAI → DBIST AI Security Institute + “staff to support the new AI Taskforce in the Office for the Prime Minister” → CO
"Rt Hon Lisa Nandy MP as Secretary of State for Digital, Culture, Media and Sport"..........
Hoping we see the announcement of the first Minister for Digital Public Infrastructure.
Deckchair shuffling is for the unserious. loosemore.com/2026/07/18/d...
Deckchair shuffling is for the unserious
The FT and Guardian report that the new PM will disband the Department of Science, Technology and Innovation (DSIT), splitting its functions between a beefed-up Department for Business and Trade (D…
loosemore.com
I only realised how thoroughly AI has raised my expectations when I tried to use a new device which still demands I learn to speak robot. (Frowns at the Home Assistant Voice Preview)
Gradually. Then suddenly. I fear it's all just too late now, very sadly.
Michael Savage@michaelsavage.bsky.social · 3w agoNEW: BBC licence fee non-payment accelerating faster than expected - fall of *more than 500k* licence fee payers in last year, bigger than internal projections. Streamers/YouTube hitting hard - big push now to include streaming in when licence is needed www.theguardian.com/media/2026/j...
Useful looking piece from the IETF on differing architectural approaches to online age verification and their differing implications for efficacy, privacy, rights &c www.ietf.org/archive/id/d...
Age Verification Architecture
This document describes solution-agnostic and technology-neutral schema for how various intermediaries can gate content and services based on age. The analysis of the architecture is done along two di...
ietf.org
I've stopped giving my blog posts a final sub, on the basis that I'd rather people think I can't spell / do grammar than think I'm getting AI to write stuff.
Just blogged on why creating LLM eval suites can be... fraught. The reasons have nothing to do with AI. loosemore.com/2026/07/09/l...
LLM eval suites are vital. And perilous.
If you’re making an AI-powered thing for real, then start by creating a comprehensive suite of evaluation test cases (aka an ‘eval suite’) with which you can continually evaluate the quality …
loosemore.com
I mean, you'd hope the NHS comms team would have at least done us poor readers the honour of removing the em-dashes. www.england.nhs.uk/blog/why-str...
NHS England » Why strategic commissioning matters now: lessons from the past, what is different this time, and what comes next
NHS England » Why strategic commissioning matters now: lessons from the past, what is different this time, and what comes next
england.nhs.uk
After chatting with a former colleague yesterday, a cynical part of me wonders if the primary outcome some senior civil servants and ministers want from a public service for which they are accountable is 'don't let it become a news story that damages my career'. Some. Not many.
Octopus Energy isn't perfect, but playing with Home Assistant has shown me just how seriously they take their API (v18.3.3 just dropped). Literally decades ahead of legacy energy companies.
So today brings the best chance ever for Count Binface to become an MP
New blog post: Prompt injection and the perils of leaving it to AI to handle AI loosemore.com/2026/07/06/p...
Prompt injection and the perils of leaving it to AI to handle AI
I’ve written recently about how citizens’ adoption of AI is putting stress on public services that rely on friction to reduce demand. I’ve since had many fascinating conversations…
loosemore.com
It is kinda mad that Stephen is beavering away on this stuff - from what I understand with not much company - and the journalists and so-called think tanks are all like, does Andy hate Reading
Stephen Abbott Pugh@stephenabbottpugh.bsky.social · last mo.Lessons so far from building opencheck.world: 1) Data quality and predictability help AI do the right thing 2) API access and documentation is crucial 3) Exposing the need for entity-resolved knowledge graphs stephenabbottpugh.medium.com/lessons-lear... #LEI #openstandards #opensource #AMLR
Pleased to see GDS recognising that the application of the Service Standard is causing problems in and of itself. gds.blog.gov.uk/2026/07/02/e...
Evolving the Service Standard for the future of public services
We’re evolving the Service Standard from a one-off test into a living standard for modern, end-to-end public services. Here’s how – and how to be involved.
gds.blog.gov.uk