Product management in the AI era: stack decisions, AI feature prioritization, business case, ROI, workflow integration, and the PM role in products that use artificial intelligence.
Today’s radar is short, but the TV4 case speaks directly to anyone designing products. An incident whose cause has not been confirmed cost a broadcaster months of work and raises an important question about how much autonomy we give automation.
Today’s story is less about a new model and more about what keeps an AI product going after it gains traction. Character.ai has filed for Chapter 11 protection to sell the company under court supervision. I looked at that story alongside seven items on the radar, from Gemini 4 Argon to Anthropic’s redesign of version control.
The question I keep coming across in my reading and product meetings is simple: what happens to the data when someone pastes something into a chatbot? I found a tool that tackles that directly, along with six items for the radar.
When an agent runs code it has just generated, the security question is not only “what if it gets something wrong?” It also matters where that error happens and what the code can reach.
Today, what held my attention most was cost per task, not capability. The same radar also brought on-device voice that runs without the cloud and a serious discussion about sandboxing code written by agents.
Today's edition is about price, not capability. On the same day, a cheaper model arrived alongside GPT-6 with an interface inside the chat and an agent configured in YAML.
The story that held my attention today was the announcement of a large open-weight model. It comes with four other reads worth your time, from agents doing science to an AI tutor that works but still struggles with adoption.
Today's radar brought the announcement of a 501-billion-parameter open-weight model, a search API for agents, and an AI case with a reputation problem. But what stayed with me was a study of a math tutor. I'll start there.
The week began with three frontier models competing on price. But the thing that made me think most was an automation tool. I’ll start there and leave the models for the roundup below.
Five stories made it through my filter today. For product teams, the biggest one is OpenAI's Dots agent. The important shift is not only what it can do; it is what changes in product design when an assistant stops waiting for a command at every step.
Six stories made it through my filter today. For product teams, the biggest one is OpenAI's launch that shortens the distance between a hypothesis and something clickable. I'll start there, then share the rest of the radar.
This week, the conversation around AI agents caught my attention more than the stream of new models. Here’s the lead story, followed by the full roundup I filtered for you.
A context window of up to 1 million tokens could change workflows built around contracts, statements, policies, and reports. Long context only pays off when it improves the work without making cost and latency impractical.
With 63 models released in September, the product decision is no longer which model to choose once and for all. It is to build architecture that lets you switch without rewriting the feature, and evaluations that show whether a new model improves your workflow.
Sonnet 5.5 keeps Sonnet 5's per-token prices but uses fewer tokens to complete the same task. Anthropic says that can reduce task costs by up to 30%. The model also generates output more than 30% faster and comes close to Opus 5.5 on several knowledge-work tasks. For product teams, this can reopen automations whose economics did not work before.
When AI reduces the time needed to deliver a service, billing only for the hours worked can put margins under pressure. The question becomes what the customer is actually paying for — and how the price should reflect that value.
This week’s radar brought near-simultaneous launches from four major AI labs and a new agent tool from NVIDIA. But the story that caught my attention was smaller: a generative AI builder aimed specifically at finance. That shift toward vertical AI builders is the thread I chose to explore today.
Today’s radar is heavy on architecture decisions. There’s a cheaper model, fresh warnings about agent security, and a new option from OpenAI for teams building AI products. That’s the thread I want to pull.
There has been a model launch almost every day this week — GPT-6 Sol and Luna, Claude Opus 5.5 and Google's new text-to-speech models. But the case that held my attention is not a recent launch. It is the way Stripe chose to automate work inside the company first.
Today’s radar was packed: three model launches on the same day — GPT-6, Claude Opus 5.5 and Gemini’s TTS — and, in the middle of it, a real case of applying AI inside a serious company. I am sticking with the second story. That is where the difference lies between a polished demo and a product that can withstand an audit.
There was a lot on today’s radar: a new Anthropic model, Grok inside Tesla cars and two serious cases showing what happens when we trust AI too much. In the middle of it all, the story that stayed with me was a small project that speaks directly to people building products in regulated industries.
Perplexity’s Hybrid Compute adapts where processing happens to the sensitivity of the information. A task starts in the cloud but can move to the user’s computer when it touches personal data. For teams building financial products, it is a direct example of privacy by design and hybrid AI architecture.
StepFun's Step 5 Preview combines a 1-million-token context window, 27 billion active parameters and a price of about US$1 per million input tokens. For product teams, the point is not the technical hype: it is how AI inference cost changes the economics of which long-running automations can make sense in production.
AI chatbots are becoming more persuasive than humans in opinion conversations. For product teams, that can reduce friction and help people move through complex journeys — but it also increases responsibility for how that influence is used, especially when the decision involves money.
Today’s radar had everything from an AI that fits in 8 MB to one that is already leading part of its creator’s own R&D. In the middle of it, one story made me stop and think about how we prove that AI can be trusted in sectors like mine.
Astra for Law shows how to turn a strong model into an AI product for regulated markets: combine general intelligence with proprietary data, compliance and the workflow customers already use. For anyone working on credit products, that choice says more than any isolated model benchmark.
Today I set aside a small study that unsettled one of my assumptions: that the most sophisticated tool is always the best choice. The radar also includes GPT-6 Astra and Nvidia’s acquisition of Hugging Face, but I started with the item that most changed the way I think about product.
This was one of those weeks when the volume of AI launches almost overwhelms anyone who has to decide what belongs on the roadmap. I pulled out the points that really matter to product teams—and today’s thread is about cost, not intelligence.
This week brought a new model, a billion-dollar acquisition and another case of an AI agent going off the rails. But what stayed with me was a clause hidden in the terms of use for an automation tool. I put the rest of the radar right after it.
Today’s radar brought a dense mix: a new frontier model, a billion-dollar acquisition and another case of an AI agent going off the rails. I kept thinking mainly about the story that combines big money with concentration risk: Nvidia’s acquisition of Hugging Face.
This week had a full menu: a new model with a critical-risk label, a billion-dollar acquisition reshaping the map of open infrastructure, and a different way of thinking about AI architecture that caught my attention. I selected what matters for people who decide product roadmaps.
Today’s text started with a detail that almost got lost in the excitement around [GPT-6 Astra](https://openai.com/index/gpt-6-astra/): OpenAI itself chose to call the model’s cybersecurity capabilities “critical”. That is what I kept thinking about all week, and it is the thread running through today’s edition.
This week was noisier than usual: a new model, a $12.9 billion acquisition, three-times-faster inference—and an AI agent that went off the rails without anyone knowing for months. I chose to follow that last thread.
This week brought a new model, a billion-dollar acquisition and even an AI agent that got out of control. But the thing that stayed with me was a much quieter tool: a simple way to prove that a file was made with AI.
This week’s radar was packed with launch after launch—Meta, Google, Anthropic and even a European alternative entering the race for frontier models. In the middle of it all, one question became clearer than ever for anyone building an AI product: which model makes sense for each task, and at what cost?
It was a day of launches in quick succession: Meta, Google, Anthropic and a European player all arrived with a new model at nearly the same time. In the middle of it all, I pulled together one read that helps make sense of so many options—and nine more links from today’s radar.
It was a big launch week for Anthropic, but the number that caught my attention most was not about model quality. It was about security. I pulled together that story and nine more links worth clicking for anyone working on AI products.
Today’s radar brought plenty for anyone following applied AI in products: from an announcement that redraws where AI lives inside the CRM to a strong warning about autonomy without a sandbox. I pulled together what matters most for people working on products, starting with the story that caught my attention.
Today’s radar revolved around money: who charges how much per token and why that is changing so quickly. The day’s highlight follows that thread through an Anthropic decision, while the rest of the curation brought plenty of agent launches to complete the picture.
One story is worth more than the other nine combined this week: an AI supplier cutting off a third-party product overnight. Alongside that came a leadership change at Google DeepMind and another chapter in the price war between models. But the most important lesson is about supplier dependence.
Today’s radar brought a heavy package of new models — Gemini, transcription, multimodal capabilities — but what stayed with me was a launch outside the model world: an agent taking over an entire sales team’s CRM.
Today’s radar brought plenty of new model launches, but another story stayed with me: a multibillion-dollar negotiation that could reshape who controls open AI infrastructure. Let’s get straight to the point.
Today’s radar was full of model launches — GLM-5.3-Flash, Qwen3.8-Flash-Next and the mystery of Ox Alpha solved. But what stayed with me was another, quieter story that is closer to my day-to-day product work.
This week brought plenty of news about chips, pricing and new models, but what held my attention most was an essay about designing freedom for AI agents. I share below why it changed the way I think about automation in credit, followed by the rest of today’s radar.
Today’s radar revolved around money and data: Nvidia backing an open model to rival China, Hugging Face reportedly valued at US$13 billion, and Thomson Reuters showing what it costs — and what it is worth — to build a proprietary model. I picked out what matters for people who work in product.
Today’s radar brought a lesson that applies to every product: being technically best does not guarantee adoption. Many developers consider Claude the most capable model available, yet cheaper alternatives are gaining ground. The day also brought plenty of movement from major players entering new spaces, from enterprise implementation to coding agents.
This week, a study about AI and school tests made me rethink how I read adoption dashboards at work. The rest of the radar ranges from the MCP roadmap to a browser built specifically for agents.
This week I spent some time reviewing the cost of running AI in the products we build, and an OpenAI price cut pulled the thread for today’s edition. The rest of the radar ranges from sub-50ms TTS latency to a billing bug in a coding agent.
The week closes with a number that captures the moment well: AI in product has stopped being an experiment and become an investment line expected to deliver results. Today’s radar ranges from age segmentation in chat to a billing bug in a coding agent.
This week’s radar brought one of those rare moves: direct competitors agreeing on a common standard. The rest of the day followed the same pattern, with payment infrastructure consolidation and increasingly cheaper models.
This was another week full of model launches and price competition, but what stayed with me was a number in the billions changing size. I will focus on that in today’s lead story, and leave the rest of the radar right after it.
I closed out the week watching an acquisition that says more about the AI market than any model launch. And the rest of today’s radar brought an interesting counterpoint: while one company is paying billions for infrastructure, the debate on Hacker News is about whether models are actually improving or merely seeming different.
Today’s radar was packed with model launches—Qwen, Gemini, GLM—but what stayed with me was an infrastructure story, not a new model. I set this edition aside to talk about homomorphic encryption and why it matters to anyone working with sensitive data in credit.
Today’s radar brought a wave of model launches—Grok, DeepSeek, Qwen—but what stayed with me was a much simpler discussion: how do we trust AI-generated text? I set today’s edition aside to talk about watermarks and document fraud, because that is what determines whether a financial product can withstand real pressure.
One data point stuck with me this week: 43% of SaaS companies already charge for work performed, not access. The rest of the day brought launch after launch—Gemini reaching 1 billion users, Alibaba with a new model, Meta entering the coding-agent race. But I set today's edition aside to talk about money, because money is what decides what survives.
Today's thread was capital chasing compute. I saw a story that put two worlds I follow closely side by side—structured credit and AI infrastructure—and decided to pull on that thread first. The rest of what passed the filter is below.
The thread today was money moving toward proven results, not promises. I started there and left the rest of what passed the filter below.
The thread today is price: a cheaper model becoming a strategic decision, not just another line item in a spreadsheet. I started with what this changes in the roadmap and left the rest of the radar below.
Today’s radar brought a bit of everything: a cheaper model putting pressure on prices, a real incident involving an autonomous agent that went out of control, and a new way to look at the AI budget. I want to start with the last one.
Today's radar has fewer flashy model launches and more of a calculation that every AI PM will soon need to make: how much does a price increase from your favorite model API weigh on your product?
The data that held my attention today did not come from a model launch. It came from people getting AI-agent approvals wrong. That weighs more on my product work in credit than any new feature of the week.
Today was a day for opening two different drawers. One was price, with OpenAI cutting Luna's entry cost by 80%. The other was people, with Jeff Dean and three other strong names leaving Google on the same day. I started with a cost spreadsheet and ended up looking at a supplier's org chart.
Today's radar was heavy on money: a wallet for an agent, token prices down 80%, and a frontier model fitting on a single GPU. I started the day thinking about spending limits and ended it thinking about unit economics.
Today’s radar was packed with model launches, but the item that stopped me was an essay about measurement. I will start with it, then share the rest of what passed through the filter.
The day brought plenty of news, but one story has stayed on my mind since early on: a major bet being shut down in less than a year. I will start there, then cover the rest of what crossed my radar.
Sunday, ten items on the radar, and one of them stayed on my mind all day. It was not a model launch, but a small change on a settings page. I will start there.
I closed the month by looking back: July was the noisiest month of model launches I can remember. Ten items made today’s radar, but I started with the one that explains all the others.
Ten items are on today’s radar, and one clear thread runs through almost all of them: what used to be an engineering project is becoming an off-the-shelf offering. I started with the launch that changed my priority list the most.
Ten items are on today’s radar, and one challenged an assumption I had been carrying without questioning it. The rest of the list points in the same direction: fewer model announcements, more discussion about how to sustain what has already reached production.
Ten items are on today’s radar, and one changes the economics of several projects that had been stalled. The rest of the list is less about new models and more about how to measure whether the work is actually delivering results.
It is Monday and the radar is already full: ten items filtered, most of them about plumbing rather than capability. New models have become routine. What really changed this week was the layer that determines whether an agent leaves the pilot stage.
Today’s radar was dominated by one theme: what happens after AI enters the delivery pipeline. There are model launches and agent infrastructure becoming a commodity, but what made me pause was the question of why measured productivity does not turn into product delivered.
Today’s radar had one dominant theme: agents are leaving the pilot stage and running into identity, permissions and auditability. I share my take first, followed by everything else that passed the filter.
Today the radar is heavy on costs and agents. GPT-5.6 arrived in three versions in one move, with Grok and Meta launching on the same day, and the result is another drop in the price of running AI. I cover what this changes for people deciding the roadmap and then everything else that passed the filter.
Today the radar is full of model launches, but what caught my attention was not raw capability. It was the strategy behind them. I cover the main story first, then everything else that passed the filter.
I am opening today's edition with a number that matters to any team considering putting AI agents to work for real. Then I cover the rest of what crossed the radar: new models, desktop agents, and a study on chatbot manipulation that is worth reading.
Today's radar is packed with model launches — Kimi K3, LM Studio Bionic, and Step 3.7 Flash, which sparked today's main story. Alongside them, I also picked out a study on dark patterns in chatbots and an account of the fatigue felt by people reviewing agent work. The full radar is worth your time below.
Today's radar brought plenty of model launches — Sonnet 5, GPT-5.6, Kimi K3 — but what held my attention was a quieter discussion: how much can human review still hold together when the volume of agents multiplies? The rest of today's curation is right below.
There were plenty of model launches in today's radar, but something else held my attention: the moment an AI agent stops answering questions and starts planning its own work process. That is where I want to begin.
There were plenty of model launches for today's radar, but what stayed with me was a study about how AI chatbots talk to us. It is worth reading before looking at the rest.
Today's thread was pricing. Sonnet 5, GPT-5.6, and Muse Spark arrived with tiered price tables, while a new report points to where the economics are heading: fewer fixed subscriptions and more outcome-based billing.
It was a busy day on the AI-for-product radar: from an agent buying stock on its own to another inference chip entering the market. But one story tied everything together—agent autonomy making its own decisions with real money on the line.
Today brought a new model with almost every piece of news, but the one that caught my attention most was the smallest of all. A model that runs directly in your pocket without relying on the cloud. I have made it the focus of this edition, followed by the rest of what I filtered below.
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.
Today's radar brought the usual: a model launch, a delayed launch, and another AI feature rolled back over privacy. But the item that held my attention most was not about parameters or benchmarks. It was about how to organize work when part of it is done by an agent.
Today brought a large wave of model launches, but what stayed with me was a quieter move: putting an agent inside the product people already use instead of creating yet another new screen. That is the thread running through today's edition.
Today, some people were cutting AI costs in half while others were being sued for stealing its secrets. It is almost a perfect summary of the phase we are in: everything is moving fast, including the disputes.
Today’s radar was packed with major model launches — GPT-5.6, ChatGPT Work, and Muse Spark 1.1. But what really caught my attention was a much quieter benchmark about accounting.
This week had more frontier model launches than you can properly keep up with — Grok 4.5, GPT-Live, GPT-5.6. But what kept rattling around in my head today wasn't any of those. It was a much simpler question about price.
Today's radar brought a bit of everything: a frontier model on the way, Meta going strong into generative media, and a security alert every agentic product PM should read. But what caught my attention most was an operations case, not a new model.
Today's radar was lighter on launches and heavier on real data about what AI is doing to the labor market. And the number I picked for today's highlight is worth reading in full.
Today's throughline was clear: in AI, the model is rarely the bottleneck. Microsoft just bet billions on this thesis, and much of the rest of today's radar confirms the same pattern, from agent costs to governance.
Today's radar came packed with contradictory signals: on one side, real data on the cost of scaling coding agents; on the other, reports of regression in the very tools we use every day. It paints a picture of a moment where technology moves fast, but not always forward.
The PM in AI products needs to understand technical trade-offs (model, cost, latency), translate technical capability into business value, design governance and the human experience around AI, and build the business case. The model is one piece — the product is the whole, including workflow, data, people, and process.
Prioritize by value generated vs delivery and operational cost. A feature that saves ten hours of manual work per week is worth more than the cents in tokens. Also consider governance maturity, technical feasibility, and alignment with existing workflow. AI features that require process redesign cost more to deliver.
Figma published a map with 13 AI tools organized by product lifecycle stage. The point is not to know all of them, but to understand which problems each category solves: research, ideation, prototyping, analysis, documentation. Knowing the landscape helps make better stack decisions.