This week brought a new model launch almost every day, but what caught my attention most was not a benchmark. It was a simple example of a B2B product replacing a complex screen with a conversation. I have captured it below, together with the rest of today’s filtered radar.
Whenever I take part in a discussion about reporting and dashboards in a B2B product, someone eventually asks: why does the user still need to learn to navigate fifteen screens to find an answer they could simply ask for? This week, Google provided a practical answer.
They launched Ask Ad Manager, an agent built with Gemini inside Google Ad Manager. The idea is straightforward: the publisher writes what they want to know in natural language, and the agent pulls together account data, builds the report, and even helps them navigate the platform itself. No learning filters and no memorizing menu paths.
What stands out is not the advertising product itself, but the pattern behind it: take a B2B system full of manual reporting and complex screens, and replace that with an agent that answers questions using the customer’s own data. This can be replicated in any vertical with data-dense operations, and structured credit is one of them.
In credit product work, I still see many people treating “generate a portfolio report” or “explain a receivables metric” as tasks that require training the user on the tool. The path Ask Ad Manager demonstrates is different: the system understands the question, retrieves the right data, and returns an answer without requiring someone to become an interface specialist.
This does not replace people who understand the business. On the contrary, it frees them from the operational work of digging up data and leaves them more time to interpret the result and decide the next step. AI handles the tedious work so the team can focus on the work that matters.
I think cases like this are worth more than any new-model benchmark because they show AI solving a real product problem within an existing flow that simply became easier.
For anyone who wants the launch details, I have included the link here: Read more
The rest of the radar
GPT-5.6, OpenAI’s next generation of models — Redefines the cost-benefit benchmark for AI features and makes “ultra” mode (parallel agents) a new product standard. Read more
Meta launches Muse Spark 1.1, a multimodal agentic model — Opens a public API for a model strong in computer use and multi-agent orchestration, expanding vendor options beyond OpenAI, Anthropic, and Google. Read more
OpenAI launches GPT-Live-1, full-duplex voice for ChatGPT — Signals OpenAI’s bet on voice as a primary interface for long-running agentic work, relevant to anyone with voice on the roadmap. Read more
Study of Claude Code and Copilot CLI adoption at Microsoft — Real-world data on internal adoption of coding agents at scale, useful for calibrating rollout expectations and success metrics. Read more
AI-agent maturity benchmark for engineering teams — A quick framework for assessing how ready a team is to adopt agents in the development workflow. Read more
Sx 2.0: sharing AI skills through a Dropbox folder — Reduces the friction of distributing standardized prompts and capabilities across a team without dedicated infrastructure. Read more
Clawk: disposable Linux VMs for coding agents — Isolation infrastructure that becomes central to security and reliability when granting coding agents more autonomy in production. Read more
“What will be left for us to work on?” — An essay on how AI automation is reshaping work roles, with direct implications for AI product positioning. Read more
That is what I set aside today. More tomorrow.