In partnership with
Hello,
This is Simon with the latest edition of The Weekly. In these updates, I share key AI related stories from this week's news, list upcoming events, and share any longer form articles posted on the website.
P.S. — sharing this issue with a couple of people gets you a free guide. More on that partway down.
I overheard a conversation recently about how simply asking chatbots basic questions doesn’t get anywhere near the full potential of modern AI tools. If I think back to when I started writing Plain AI well over a year ago, that was certainly how I used AI most of the time. I had a question, I typed it in as a prompt, and I waited for the answer. Sometimes that was enough, and other times I would have a few follow-up questions. Fast forward to today, however, and AI chatbots are far more capable. The conversation I overheard went on to highlight that the real value these days is in getting AI to do something for you.
This is something I can really agree with. Just this week, I realised that I spend quite a lot of time manually looking through calendars to find an overlapping gap to slot in a meeting. As this is a repeatable task, I decided to get AI to help and created a Skill in Claude. What’s really neat about it, and a good example of AI doing something for you, is that it puts three placeholders in the other attendees’ calendars while I coordinate the actual availability over email. This use case isn’t going to change my world, but it saves me 10 minutes of pain that I go through 3-4 times a week, so it’s well worth it.
But this is just a fraction of what AI can do for us. Over the course of a week, I’ll ask Claude to create a presentation, produce a PDF or help me code something on my coffee website. I also have scheduled tasks that run entirely on their own to help me with research. At this point, it really is just our own imagination that limits us. The technology is certainly there, should we choose to use it.
Beyond basic questions, what are you getting AI to do for you?
A Message From Our Sponsor
Blu Dot surpasses 2,000% ROAS with self-serve CTV ads
Home furniture brand Blu Dot blew up on CTV with help from Roku Ads Manager. Here’s how:
After a test campaign reached 211,000 households and achieved 1,010% ROAS, the brand went all in to promote its annual sales event. It removed age and income constraints to expand reach and shifted budget to custom audiences and retargeting, where intent was strongest.
The results speak for themselves. As Blu Dot increased their investment by 10x, ROAS jumped to 2,308% and more page-view conversions surpassed 50,000.
“For CTV campaigns, Roku has been a top performer,” said Claire Folkestad, Paid Media Strategist, Blu Dot. “Comping to our other platforms, we have seen really strong ROAS… and highly efficient CPMs, lower than any other CTV partner we've worked with.”
Using Roku Ads Manager, the campaign moved from a pilot to a permanent performance engine for the brand.
Real World AI Success Story
UPS's ORION route-optimisation system saves 10 million gallons of fuel and an estimated $300–400 million a year by recalculating delivery routes for 55,000 drivers nightly — a project that took a decade to build and won the top prize in operations research in 2016.
UPS began deploying ORION — On-Road Integrated Optimization and Navigation — across its US network in 2012, completing the full rollout of 55,000 drivers by 2016. The system uses operations research and machine learning to compute optimised delivery sequences, eliminating left turns where possible to reduce idling time and fuel burn, and incorporating real-time inputs including traffic, weather and customer-requested time windows. The results, verified independently and published in the INFORMS journal that awarded ORION the 2016 Franz Edelman Prize for achievement in operations research: 10 million gallons of fuel saved annually, 100 million fewer miles driven per year, and cost savings estimated at between $300 million and $400 million per year depending on the fuel cost assumption used. UPS's own 2018 sustainability reporting confirmed the CO₂ reduction at around 100,000 metric tonnes annually. The system has continued to evolve: UPS added a dynamic re-routing layer in 2018 that allows route adjustments mid-day as packages are added or time windows change, estimated to add a further 100 million fewer miles per year on top of the original savings. ORION is the rare enterprise AI story where the headline figures have been confirmed by an independent academic body, published in peer-reviewed context, and held up across multiple annual reporting cycles.
Turn your meeting notes into answers — free
If this issue was useful, share Plain AI with a couple of people who'd get value from it too. When two people subscribe using your link, I'll send you my exclusive guide, "Supercharge Your Work with NotebookLM" — a practical, no-nonsense way to stop drowning in meeting notes, reports, and research, and start actually using them.
Curated AI News
OpenAI starts watermarking ChatGPT text in the EU
OpenAI will embed invisible, machine-readable watermarks in text generated by ChatGPT and Codex for EU users to comply with the AI Act's transparency rules, which took effect on 2 August. It is available as an opt-in API feature from 5 October and becomes mandatory for eligible EU output within weeks, according to Gizmodo and AI Weekly. TechCrunch reports detection accuracy falls from about 92% to 66% if roughly 10% of words are swapped, and a missing watermark does not prove human authorship. Anthropic introduced similar marking in August.
Why it matters: The first visible effect of the AI Act on day-to-day tools is that identical prompts may produce different output in Europe. If you operate across borders, check how your teams, contracts and content policies handle AI-generated text, and don't treat watermark detection as proof of anything.
Who pays when an AI agent goes wrong?
Reported incidents in which AI agents hacked third-party systems during tests, including OpenAI agents that breached Hugging Face and hijacked a German wiki site and RubyGems, have exposed gaps in liability law, according to MIT Technology Review and CNN. Current US state safety laws only require reporting of incidents causing 50 or more deaths or injuries, or $1bn in damage. State attorneys general in Alabama, California and a Montana-led coalition of 16 states are asking questions, and Senator Josh Hawley has opened an investigation. Hacking statutes struggle with whether an agent can have "intent."
Why it matters: If you deploy AI agents with access to your systems or customers, assume responsibility lands on you rather than the supplier until the law says otherwise. Check what your contracts and insurance actually cover.
Microsoft reportedly paying Anthropic $2bn+ a year to fix Copilot
Microsoft is on track to pay Anthropic more than $2bn a year to power parts of Copilot, according to The Information. Internal pressure reportedly grew after customers complained Copilot wasn't good enough at automating work in Excel and PowerPoint, and Anthropic's models performed better on those tasks. Microsoft is blending Claude with OpenAI models across Word, Excel, Outlook and PowerPoint. The figure comes from a single unnamed source and has not been confirmed.
Why it matters: Customers' unhappiness with Copilot is the real signal: even the biggest vendors are still working out whether AI actually does the job in everyday office tools. Judge your own rollout on whether tasks get done, not on which logo is on the licence.
Upcoming AI Events
Data & AI Conference Europe
London, UK, 13 – 17 OctoberFT - Future of AI
London, UK, 5-6 NovemberBig Data Conference Europe
Vilnius, Lithuania, November 19-21AI World Congress
The Great Hall, London. November 27-28
Thanks for reading, and see you next Thursday.
Simon,
Was this email shared with you? If so subscribe here to get your own edition every Thursday.
You can also follow me on social media: