Thought #16 - The AI Shake-Up Continues

Hi lovely humans,

It’s been another busy week in AI, a slightly different week without big flashy releases but lots of small impactful changes (for example, Anthropic making more things available on free and cheaper packages). But the big news this week is Keir Starmer's opening at London Tech Week - we're going all in on AI.

As part of that, the government announced a new national skills drive, aiming to boost AI and tech education - especially for young people. It’s light on detail so far, but does name Amazon and Microsoft as partners already offering training. While that’s a positive sign, we believe the real focus needs to be on how we upskill the current workforce - not just future talent. AI is changing jobs now, and the support needs to match that pace.

It’s a long one this week - lots of little important things going on. Again, a very quiet week from OpenAI.

What We’ve Been Up To

Being Part of the Claude Fan Club

I’m a bit obsessed with Claude at the minute - especially the new integrations I can access on Pro. In particular the Atlassian one - which let’s me use JIRA in my Claude chats. JIRA tickets are what developers use to plan work, and being better at actually using these is becoming important to me as we’re hiring a full stack dev and an intern.

The tickets are very high quality:

Claude has also been helping me plan and prioritise our next month’s dev work, and seems to be giving realistic estimates. Apparently the next step is to get Claude to do all the coding, but I am not quite there yet.

AI New Releases

Subtle but powerful updates that matter for workplace tools

ChatGPT Business gets serious

OpenAI has rolled out business-focused updates, including new o3-powered workflows and integrations with Google Drive, HubSpot, Dropbox and more.

Claude keeps evolving

Claude Code is now available to Pro users (this is Claude in your terminal actually coding), while Claude Projects supports 10x more content. A new voice mode allows Claude to query your Google Calendar or Gmail out loud - bringing it closer to a true assistant. Meanwhile, a new blog on coding with Claude quietly launched with some intriguing prompts.

Apple’s AI “catch-up”

At WWDC, Apple announced a Shortcuts app powered by AI - and reiterated plans to finally improve Siri. Critics pointed out they still feel behind the curve, but the integration across devices could matter longer-term. TechCrunch weighed in here.

Google pushes Gemini Pro 2.5

New features include scheduling and voice-to-text across 24 languages. They also quietly launched an Android app to run open-source AI models locally.

Mistral and Perplexity both ship updates

Mistral launched a code model and is reportedly raising $1B.
Perplexity added memory, topic categories, SEC data access, the o3 model and did a Reddit AMA.

AI News

Government strategy, job shifts, and governance questions

UK Government doubles down on AI

Announced during London Tech Week, the UK launched a set of AI-related initiatives:

• A digital tools marketplace aiming to save £1.2B

• AI-driven planning permission to speed up 1.5 million homes

• A national skills drive focusing on AI and tech, mostly aimed at young people

• Ambitions to become a global leader in AI drug discovery

Entry-level jobs under threat

In a widely shared interview, Anthropic’s CEO warned that AI could replace “half of all entry-level jobs”. It’s a stark warning that echoes wider concern about who is being left behind as AI adoption accelerates.

AI job titles rise, entry roles fall

A report from Ravio found that job titles containing “AI” have increased 7x in 2025 so far. At the same time, entry-level tech hiring has dropped by 73.4%, showing how uneven the impact of AI is on different parts of the workforce.

OpenAI publishes guidance on vulnerabilities

OpenAI released its position on how it handles third-party vulnerabilities discovered through its models. It’s a small but important move toward more responsible model deployment.

Nvidia becomes most valuable company

Nvidia has overtaken Microsoft to become the most valuable company in the world, reflecting just how central its chips are to the AI boom.

FCA teams up with NVIDIA on AI

The UK’s financial regulator is creating a sandbox environment for firms to experiment with AI tools alongside NVIDIA. It’s a notable sign that regulators are shifting from observation to active facilitation.

Samsung may integrate Perplexity into its phones

Samsung is reportedly considering a partnership with Perplexity to build its AI assistant tech directly into future smartphones. It’s an early-stage move, but one that would put Perplexity in front of millions of users - and signal growing competition in the AI assistant space.

Protein-binding breakthrough

Researchers have developed Boltz 2, a new model that predicts how strongly molecules bind together. It’s a major step forward in AI’s role in drug discovery and biomedical science.

Largest known data leak in China

A breach uncovered in China exposed billions of personal records, in what’s being called the largest known data leak ever. It raises fresh concerns about data security in an AI-driven world.

AI in education - still vague

The BBC ran a first-look piece on how teachers could use AI to save time on marketing. But the examples were thin, and most educators on LinkedIn called for proper support and training over vague promises.

AI misuse in UK court

A UK judge issued a warning after AI-generated fake legal citations were submitted in court. A reminder that misuse isn’t just theoretical - it’s happening.

Meta builds AGI team

Meta is hiring a team focused on developing Artificial General Intelligence (AGI). One to watch, though few details have emerged so far.

Anthropic adds Reed Hastings to board

Reed Hastings, Netflix co-founder and longtime tech investor, has joined the board at Anthropic. It's a sign of serious growth ambitions.

Reddit claims Anthropic still scraping data

Reddit says Anthropic has continued scraping Reddit content despite an ongoing lawsuit and previous denials. Another battle in the training data wars.

Grok misinterprets Elon Musk

Grok, the AI chatbot from xAI, mistook a phrase from Elon Musk as a murder confession. Musk had to clarify, and the moment quickly turned into internet folklore.

Not Quite News, But Worth a Read (or Listen or Watch)

• How FT defines AI fluency
  Matt Partovi shared how the Financial Times approaches AI fluency in-house - a useful reference for anyone writing internal skills frameworks.

• The six real ways to use AI
  Ben Cohen offers a tidy categorisation of AI use cases. Not definitive, but a helpful tool for team discussions or training.

• Can weather forecasters survive AI?
  A BBC veteran reflects on AI’s role in weather prediction after 30 years in the job. Gentle, thoughtful, and very relatable.

• Abba member announces AI musical
  Björn Ulvaeus revealed plans for an AI-assisted musical. It’s not really news - but it is a sign of how culturally embedded AI is becoming.

LinkedIn AI Poll

Last week we asked about using Perplexity (an AI chatbot specifically for web search and research).

43% of users have never used Perplexity! Much higher than for Google Gemini last week, but similar percentage using them sometimes.

Vote in this week’s poll - please!

This week we want to know if you’re using Microsoft Copilot.

Our Blog and Resources

I compared Deep Research from Claude, ChatGPT, Gemini and Perplexity:

We’ve been helping teams design better internal AI tools, and this checklist-style guide walks through how to test what you’re building.

Final Thoughts

As always we hope this was helpful!

Feel free to share this with anyone who might find it useful.

Next week, we’ll be here with the news and hopefully a few blogs with the bots battling each other.

Laura
Always learning

PS we’re hiring a Full Stack Dev if you know anyone!