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.

A recent topic catching my eye is the idea that with the advancement of agentic AI, we may no longer need SaaS (Software as a Service). As someone who's worked at SaaS companies for 15+ years, I find this to be both an interesting and concerning idea.

When we consider that we use software to achieve specific tasks and different use-cases, it does make sense that AI could replicate large parts of what we do.

There are two good examples of where this could happen quite easily:

Whilst we're not quite at this fully automated stage yet, it can only be a short time before this seems normal.

Here's a few other examples:

What systems do you currently use in your day-to-day job that could soon be replaced with AI?

Do you think that's a good thing or a threat to your role?

Microsoft's launch of Deep Research in Azure AI Foundry represents a significant leap forward in making sophisticated AI research capabilities accessible to enterprise developers. This isn't just another API release; it's a comprehensive research orchestration platform demonstrating how agentic systems can handle complex, multi-step research tasks that previously required extensive human coordination.

Why it matters: The system's five-stage architecture showcases the maturation of agentic AI: intent clarification, web grounding, deep research execution, transparency controls, and programmatic integration. Most significantly, Microsoft's pricing at $10 per million input tokens positions Deep Research as a premium tool for high-value research workflows rather than casual information gathering. This pricing strategy reflects the growing recognition that sophisticated AI agents deliver measurable business value that justifies premium costs.

Perhaps the most significant paradigm shift comes from NVIDIA's research demonstrating that small language models (SLMs) under 10 billion parameters can match older large language models in reasoning and code generation, offering 10-30x cost reductions. This research challenges the prevailing "bigger is better" mentality and suggests a fundamental architectural shift toward specialised, efficient models for agentic systems.

Why it matters: The implications are profound: SLMs enable real-time agentic responses at scale while dramatically reducing infrastructure costs. The paper's proposal for heterogeneous agent systems, where SLMs handle 70-80% of tasks and only selectively call large language models for complex reasoning, represents a more economical and sustainable approach to agentic AI deployment.

Capital One's deployment of a production-grade multi-agent system for its auto business provides concrete evidence that agentic AI can deliver significant business value at scale. Their four-agent architecture, customer communication, action planning, evaluation, and validation agents, demonstrates how financial institutions balance autonomous decision-making with regulatory compliance.

Why it matters: The performance metrics are compelling: customer engagement improved by up to 55% in some cases, with dealers identifying more serious leads through natural, 24/7 agent interactions. This case study proves that well-designed agentic systems can enhance rather than replace human expertise whilst delivering measurable business outcomes.

Thanks for reading, and see you next Friday.

Simon,

Was this email shared with you? If so, subscribe here to get your own edition every Friday.

Are you looking for a job in the tech sector? You'd do well to check out all the available roles on Wellfound. Wellfound is a platform that connects startups and tech companies with job seekers. It offers over 130,000 remote and local job listings, allowing candidates to apply with a single profile and view salary and equity details upfront.

You can also follow me on social media: