Andy's recaps are always fun! My takeaway from this is that SQL is (still) not going anywhere and if you're an OLAP or NoSQL database technology that doesn't support SQL, your days may be numbered

Be vigilant when buying books for your loved ones during the holiday season. Counterfeiters are also getting into AI engineering and LLMs!

You will be visited by three spirits

My thoughts on IBM acquiring Confluent (coming at you from Mountain View)

Post-AI companies that go from zero to AI may eventually find out they didn’t add enough engineering rigour. Pre-AI companies that go from product to AI product may eventually find out they didn’t have the right foundations

🗽 We are live at the Real-Time Data AI Hackathon with Lenses.io x Amazon Web Services (AWS) x Ness Digital Engineering, enjoying the creativity and views.

Nick Aldrich from @awscloud.bsky.social showing us what's next for async tool calls in MCP (though I think an event-driven agent approach with ephemeral clients is the way). There will be improvements in client-polling. Follow and contribute to it in SEP-1391 github.com/modelcontext... #MCPDevSummit

Full house at the #MCPDevSummit morning keynotes. Great seeing David Soria Parra from @anthropic.com telling us the story so far: 10,000+ integrations built by the community in the 10 months since the MCP launch.

🫡 I salute @workos.bsky.social for providing secure auth for MCP servers and also their commitment to swag at #MCPDevSummit

I'm heading to the MCP Developers Summit tomorrow. Who else will I see there? Really looking forward to the keynotes, spec updates (OAuth in particular) and war stories from MCP in production. Also quietly hoping for swag of this magnitude 🤞🏽 @mcpdevsummit.ai

You’re never alone at #PyData. Love this little group we’ve built up and stoked that all the talks today are from the local community. Kicking off the end of summer meetup in an hour. Looking forward to catching up with the regulars and welcoming the newcomers to Python Data Land 🤙🏽

Enjoyed a lovely pint on Roger Daltrey’s estate (it absolutely pissed it down 2 mins later)

Copilot support in Excel has been added by @microsoft.com. The joke is that most of a company’s data is in Excel, so how can you do the most damage to a company’s financial reporting? You enable vibe spreadsheeting 🔥

🍕TIL about the Pentagon Pizza Index, which is based on the theory that a spike in late-night pizza orders near the US intelligence HQ could signal an impending national security crisis or military action. 🚨 Currently at DEFCON 3: increase in force readiness

I've been a busy bee building an MCP server at lenses.io and just got my ticket for @mcpdevsummit.ai in October 🙌🏽 Who else is going and who should I meet?

It’s a marathon, not a sprint. This is a well known phrase, especially if you’re a founder or if you work in a startup. It absolutely is a marathon and not a sprint, but our concept of a good finishing time is broad.

There’s an influx of new AI experts in the market. AI may be slowly replacing jobs but it's fast replacing bike sheds (see: Parkinson's Law of Triviality)

🏖️ It’s a month to go till the end of summer PyData Cornwall meetup! Just updated the event page with the lightning talk speakers. Most of them are first time speakers so let’s make them feel welcome and powerful.

"add a tool that makes me a sandwich" is the new "sudo make me a sandwich". Here's my take on the Sandwich kxcd. My AI-coding tool of choice is Cline. My favourite feature: it can extend its own capabilities via Model Context Protocol. If it can't do something, you can just ask it to add a tool.

Vibe coding is like driving with GPS. You still need to pay attention to motorists, pedestrians and road safety. If you overuse it, you lose memory of routes you know. If you don't provide details and accept what you're given, you may end up in Kansas City, Kansas instead of Kansas City, Missouri

🏄 PyData Cornwall is back in September to close out the season. Let's get together to catch up on holiday stories and knowledge gained over the summer. 🙏🏽 Would love to grow the community so please repost and share with all your local friends curious about building with data and AI

I recently realised I refer to "Small LLMs" and "Large LLMs". This may seem like RAS (Redundant Acronym Syndrome) Syndrome, e.g. PIN Number and ATM Machine but it isn't. To me...

The perfect dessert doesn't exis... #strawberry #trifle #dessert

Could be Baader-Meinhof but I'm seeing frequent mentions of "high agency" lately. It was coined around a decade ago but now having its ageconaissance moment in the current era of agentic AI

60% of the time, it’s salami every time

The fins are back on the desk so it's The Fin Room once again. Love the stubby Almond Surfboards one and the pink Watermark Surfboards one too

🧘‍♂️ Must remember to take more breaks from the screen and stare into the horizon. Lucky to have this view ~2 mins walk away from today's desk

Today's the day of the first PyData Cornwall meetup! Prepped the event space this morning so it's all clean and tidy. Someone mentioned this is the most expensive cleaning team (comprised of data engineers), but there's no job too small when it comes to building community @pydatacornwall.bsky.social

Tap, tap, is this thing on? I hope so, as I wanted to say hi to @bengamble7.bsky.social who's speaking at the wonderful @pydatacornwall.bsky.social event next week. You aren't coming? Signup here - www.meetup.com/pydata-cornw... - DM with any questions and look forward to seeing you there!

🏴‍☠️ Just under 2 weeks to go till the first PyData Cornwall meetup! Learn how to create stories for children with generative AI and forecasting on a Raspberry Pi. ⚡ There's still one slot left for a 5 minute lightning talk so please get in touch if you'd like it!

This year’s @pydatalondon.bsky.social hosted a small Feminist AI Lounge on the unconference track, featuring a zine making station and great discussions. For future Feminist AI events in London, organized by the community, follow @feministai-ldn.bsky.social ⚡ feministai.party/events/2025-...

Delighted with these scones being Cornish rather than Devonshire #IYKYK #CreamOnTop

Final #PyData London keynote from Tony Mears on the ways that innovation dies, from his experience directing operations in the NHS, UK Space Agency and government agencies. Great insights into how innovation gets stifled by risk aversion and lack of a failure culture @pydatalondon.bsky.social

A comprehensive dive into why you should care about LLM evaluations from Lena Shakurova at #PyData London. Build datasets with SMEs. When you have an understanding then create synthetic data. Good tips for metrics and creating a 3 step evals framework @pydatalondon.bsky.social

⚡️ Lightning talks this morning at #PyData London. Loads of fun submissions; the wheel decides @pydatalondon.bsky.social

🐍📊🍺🌚

Going deep into transformer layers with @baggiponte.bsky.social explaining how to count parameters in order to calculate compute and memory requirements. Note to self: get an understanding of key-value (KV) caching in transformers #PyData London @pydatalondon.bsky.social

Zines being made in the Feminist AI Lounge hosted by @inesmontani.bsky.social - some amazing stickers making their way onto collages from some creative geniuses at #PyData London @feministai.party @pydatalondon.bsky.social

Adding GenAI to an existing working ML system seems unnecessary but @astroadamh.bsky.social goes through reasons to do it whilst navigating pitfalls, e.g. changes in model prices. Great idea exploring this through personas we’ve all encountered in the workplace #PyData @pydatalondon.bsky.social

At #PyData London @inesmontani.bsky.social confirms our suspicions that we should try not to work with PDFs directly. Do the expensive extraction upfront into a unified structured format and build all your analysis and #NLP components yourself so you can better modularise @pydatalondon.bsky.social

#NLP expert Ines is back to speak at #PyData London. Make sure you join the #FeministAI party after her talk. @inesmontani.bsky.social

Keynote from @jayalammar.bsky.social at #PyData London on tracing a path from simple token prediction to systems that can reason. These LLMs of the future will need a lot of supervised fine-tuning, reinforcement learning and maybe yet another type of positional encoding @pydatalondon.bsky.social

I had so much fun at the first day of the @pydatalondon.bsky.social conference yesterday I forgot to post about it 😉 My highlights from day one included 🧵

Never expected to see Stardew Valley alongside these tech logos in the same slide 🌳 @olenakutsenko.bsky.social showing us Kafka as the distributed log for conversations in an agentic video game simulation with Flink and OpenSearch #PyData London @pydatalondon.bsky.social

Lightning talk slots already filling up! Want to share something cool with the rest of the conference? Sign up at the @numfocus.bsky.social booth!

Kicking off #PyData London with a keynote from Leanne Fitzpatrick (Director, Data Science/AI @financialtimes.com) on grounding yourself as a data practitioner during the AI arms race. Loved the reminder of environmental/regulatory considerations when designing AI systems @pydatalondon.bsky.social

Welcome to #PyData London @pydatalondon.bsky.social

The afternoon tutorials at @pydatalondon.bsky.social had the AI engineers covered. First up was training @karpathy.bsky.social’s nanoGPT to tell knock-knock jokes followed by learning how to build vertical agents with the open source AgentOS AG2 (community fork of AutoGen) from its founder Chi Wang

Just finished a great tutorial from Samuel Joseph taking us step-by-step through creating a simple GPT model… to tell knock knock jokes. And building intuitive visualisations along the way!

Now learning about forecasting weather using time series data from Suyash Joshi. We’ve been introduced to @influxdb.com and are going to use NeuralProphet to predict weather in London—definitely starting on hard mode with London!