While you were sleeping (but not necessarily missing the GenAI train)

I mentioned in my previous edition that I had spent the first quarter of the year taking a bit of a break from all (okay, most) things Generative AI — something I called dialling down A to rediscover "I" — but also hinted that some of the developments that I missed through that period were the very tools that made getting back up to speed a whole lot easier.

One of those was the release of the 'agentic' Deep Research functionalities available to premium subscribers of ChatGPT and Google Gemini. I've since done extensive experimentation with both, including asking them to give me the low down on key developments, advancements, insights and issues in relation to Generative AI while I was POÄNGing my way away from monitors from late December through March this year.

Watching the Deep Research agents in action is fascinating, particularly in terms of how they reveal (to some degree) how they act, display curiosity and problem-solving, and list the sources of their search as they scour the Internet looking for information and gradually formulate hypotheses and then answers.

I now have a variety of reports to peruse, but the best came from Gemini Deep Research with 2.5 Pro, which I then got Gemini, OpenGPT and Claude all to make more concise summaries of using my 'Wotan Nine' formula (three key themes to emerge, with three key developments in relation to each theme). All three came up with the same key themes and mostly the same sorts of key developments in each — but all three models agreed Claude's summary was probably the most useful in finding a middle ground between Gemini's depth, technical focus and data specificity and ChatGPT's brevity, strategic/business focus and data framing.

This, apparently, is a snapshot of where the GenAI juggernaut went in Q1 2025:

1. The Rise of Agentic AI

This period marked a decisive strategic pivot from basic generative models to autonomous AI agents capable of planning and executing tasks.

Key Developments:

• Industry-wide Strategic Shift: Major players like Google, OpenAI, Amazon, Microsoft, and Salesforce all prioritized agent development, with Google releasing its Agent Development Kit (ADK) and Agent2Agent protocol, and nearly all cloud platforms offering agent-building capabilities.
• Enhanced Reasoning and Multimodal Capabilities: New models like OpenAI's o1/o3 series, Google's Gemini 2.0, and Anthropic's Claude 3.7 were specifically designed with improved reasoning, planning, and tool-use capabilities to support autonomous agents but also making multimodal processing (text, image, audio, video) a standard expectation.
• Emerging Challenges: Despite the excitement, significant limitations were revealed including capability gaps in deep contextual reasoning, security vulnerabilities (with Gartner predicting a rise in agent-related breaches), and difficulties in ensuring transparency and accountability of agent actions.

2. Infrastructure and Investment Acceleration

The period saw unprecedented capital concentration and infrastructure development, creating new competitive dynamics and barriers.

Key Developments:

• Record-Breaking Funding: Q1 2025 witnessed extraordinary investment, highlighted by OpenAI's $40 billion round (the largest private funding ever recorded), with total AI funding reaching $59.6 billion, representing 53% of all global startup funding.
• Infrastructure Arms Race: Massive infrastructure initiatives emerged, including the reported $500 billion "Project Stargate" led by the Trump administration, Oracle, OpenAI, and others, alongside other multi-billion-dollar data center expansions by tech giants.
• Market Bifurcation: Capital became increasingly concentrated in established leaders and essential infrastructure, creating formidable barriers to entry for newcomers and potentially reshaping competition around access to computing resources rather than algorithms alone.

3. Intensifying Safety, Regulatory, and Ethical Challenges

Real-world deployment of GenAI brought concrete safety failures, regulatory responses, and ethical dilemmas to the forefront.

Key Developments:

• High-Profile Safety Failures: DeepSeek R1, initially lauded for performance and cost-efficiency, was found by multiple security research teams to have severe vulnerabilities including extreme susceptibility to jailbreaking and generation of harmful content. Character.AI faced multiple lawsuits alleging serious harm to minors.
• Regulatory Solidification: The EU AI Act entered key implementation phases while US states introduced over 550 AI-related bills, creating a complex compliance landscape addressing concerns like deepfakes, algorithmic discrimination, and healthcare AI.
• Cost vs. Safety Tension: The DeepSeek R1 case highlighted a potential trade-off between development cost/efficiency and model safety, raising questions about whether achieving both accessibility and trustworthiness is possible without significant investment in safety measures.

These themes collectively represent a transition phase where GenAI moved beyond technical capabilities toward practical implementation, with the industry grappling with how to responsibly deploy increasingly powerful autonomous systems at scale.

(Summary generated by Claude AI, based on an in-depth report generated by Gemini Deep Research with 2.5 Pro)

So there you have it. I have some other useful and interesting summaries generated by Deep Research (including a fascinating round up of Ethan Mollick's posts and newsletter entries so far this year, and an in depth look at what almost every major consulting firm has been saying about AI implications for leadership and management over the quarter), all of which are quite congruent with the summary above and some of which I might tap into for the newsletter in coming weeks.

But for now, well done — consider yourself up to speed!

*

Image generated by Google's Imagen (within Gemini), given a pretty loose prompt to depict getting up to speed and not missing trains, but also being faithful to my Wotan Nine formula (count the doors on the train).