AI Threads

Questions asked by people building AI in the UK, and answered by the rest of them. A new one goes up every morning; members start their own whenever they have something to ask. Anyone can read a thread without signing in.

Anyone can read the thread. Starting a thread is for AISAT members. Sign in to join in, or apply if you have not already. It is free. Most applications are approved the same day. Around meetup weeks and holidays it can take a day or two.

Apply to join Sign in

56 threads, page 2 of 3

Nobody has found this useful yet
Question of the day4 Oct 2026
Microsoft has rebuilt Copilot around Autopilot, a long-running agent with its own identity that works through multi-step tasks without a person approving each step. Where do you draw the line on how long an agent may run unsupervised before a human checks its work?
No replies yet
Nobody has found this useful yet
Question of the day3 Oct 2026
Alibaba says it cut Qwen voice API prices by up to 95% for speech recognition and about 70% for text-to-speech. For a UK contact centre, is per-minute cost really what holds voice AI back, or is it the UK GDPR restricted transfer question over where call audio goes?
No replies yet
Nobody has found this useful yet
Question of the day2 Oct 2026
Google's new Credentials API for Gemini managed agents stores secrets and injects them at request time, so the agent uses a token it never reads. Do you treat an agent like that as an identity in its own right, with its own access reviews, or does it still sit under whoever set it up?
No replies yet
Nobody has found this useful yet
Question of the day1 Oct 2026
A US appeals court has upheld the Pentagon's designation of Anthropic as a supply chain risk over its usage restrictions, not its model quality. Could you say within an hour which of your agents run on Claude, and what would moving them actually involve?
No replies yet
Nobody has found this useful yet
Question of the day30 Sept 2026
Dataiku announced Agent Management on 24 September, a tool that lists AI agents across platforms and ranks each by risk, with general availability planned for October. Would you keep your agent register inside a vendor's tool like that, or in something your risk team owns outright?
No replies yet
Nobody has found this useful yet
Question of the day29 Sept 2026
An unsealed Microsoft memo calling AI scraping "the largest theft of labor in human history" surfaced in the New York Times copyright case this week. Does language like that from inside a major vendor change how you talk to clients or staff about the tools you deploy?
No replies yet
1 person found this useful
Question of the day12 Sept 2026
An insurance group announced enterprise-wide AI adoption driven by its own staff rather than a top-down mandate. From your experience, do employee-led champion networks actually outperform a mandate from the executive team, or do they just move slower?
1 reply
Nobody has found this useful yet
Question of the day28 Sept 2026
A rota optimiser, a contact centre productivity score and a CV screening model all count as algorithmic management, whatever the internal project was called. Have you ever discovered a tool in your organisation fell into that category after it went live, and what changed once you realised?
No replies yet
Nobody has found this useful yet
Question of the day27 Sept 2026
The Sovereign AI R&D Procurement Scheme assesses proposals in batches, with rounds closing 1 October 2026, 1 December 2026 and 1 February 2027, and no minimum turnover or cash reserve requirements. Does dropping those financial hurdles actually open the door to smaller UK suppliers, or does something else still keep them out?
No replies yet
Nobody has found this useful yet
Question of the day26 Sept 2026
Louise Haigh, the First Secretary of State, told the TUC congress on 15 September 2026 that Britain would heed warnings over AI. If UK applied-AI rules arrive through employment law and union negotiation rather than an AI act, who in your organisation needs to be in the room that currently isn't?
No replies yet
Nobody has found this useful yet
Question of the day25 Sept 2026
EU Annex III high-risk obligations have slipped from 2 August 2026 to 2 December 2027 under the AI Omnibus. Has your team kept building logging, human oversight and documentation to that standard anyway, or has the deferral quietly pushed it down the backlog?
No replies yet
Nobody has found this useful yet
Question of the day24 Sept 2026
The AI Energy Management Alliance offers data centres faster grid connections if they cut power when the grid is stressed, and DESNZ has been weighing a minimum flexibility duty for very large users in the UK. If you are sizing UK compute for 2027, how are you costing curtailment?
No replies yet
Nobody has found this useful yet
Question of the day23 Sept 2026
OpenAI published six reports of unexpected or concerning model behaviour plus its own disclosure framework on 16 September 2026, all self-selected and caught in training and evaluation rather than production. Would a vendor incident log like that carry any weight in your supplier assurance, and what would make it credible?
No replies yet
1 person found this useful
Question of the day31 Aug 2026
Beyond basic prompting, what is one specific AI workflow or architecture you've built recently that actually saved you hours of work rather than just being a fun experiment?
1 reply
Nobody has found this useful yet
Question of the day21 Sept 2026
The ICO has a consultation on AI guidance due to launch this month, alongside its programme on automated decision-making and profiling. What is the one thing you want the ICO to be clearer about before you'd let an automated decision touch a customer or a member of staff?
No replies yet
Nobody has found this useful yet
Question of the day20 Sept 2026
A month on from the EU AI Act's August milestones, UK regulators are still working through their own tracker of AI commitments. If you had to brief a non-technical UK board this week on your single biggest AI compliance exposure, what would you name?
No replies yet
Nobody has found this useful yet
Question of the day19 Sept 2026
Since 2 August 2026 the EU AI Act has required interactive systems to tell people they are talking to AI and AI-generated content to be labelled, and that duty falls on deployers too. Have you checked your EU-facing chat and synthetic media, and what did the disclosure actually look like in the product?
No replies yet
Nobody has found this useful yet
Question of the day18 Sept 2026
Artificial Analysis found GPT-6 Astra used roughly a third of Sol's tokens in the Codex coding harness but only about 10% fewer on general tasks, making it around 75% dearer per task there. What benchmark or test set do you trust when deciding which model handles which job?
No replies yet
Nobody has found this useful yet
Question of the day17 Sept 2026
UK public bodies awarded 453 AI-related contracts worth £1.4bn in 2026 up to August, which makes a £100m R&D pot look small. Does a scheme this size genuinely change who wins public sector AI work, or does the incumbent integrator still take the bulk of it?
No replies yet
Nobody has found this useful yet
Question of the day16 Sept 2026
The UK consultation on workplace monitoring technologies closes on 30 September 2026 and covers algorithmic management, productivity analytics and AI-assisted decisions about staff. If you've deployed anything of that kind, what would you actually tell government about transparency and worker voice?
No replies yet