Anna Via: What happened in AI and product — March and April 2026

Anna Via, an ML product manager at Adevinta and board member at DataForGoodBcn, publishes a recurring roundup of AI developments that matter to product teams. This installment, covering March and April 2026, is one of the denser editions in the series.

What the article covers

The roundup is organized into four areas: model updates, product shifts, enterprise adoption data, and labor market findings.

On the model side, it notes Anthropic’s Claude Opus 4.7 with its one-million-token context window, OpenAI’s GPT-5.5 described as “highly autonomous,” and Google’s open-source Gemma 4. These are treated not as marketing milestones but as practical signals about what product teams can now realistically build on top of.

The product section focuses on the rise of agentic commerce — AI as a primary interface for product discovery rather than a search enhancement. Via describes Perplexity’s Personal Computer as a notable shift: a product that manages objectives on a user’s behalf rather than responding to direct queries. Alongside this, she cites data showing that 84 percent of European consumers now use AI daily for shopping comparisons.

Enterprise adoption figures come from two sources. Nearly 30 percent of Fortune 500 companies are paying customers of AI startups. At the same time, only one in five companies currently has a mature governance model for autonomous agents — a gap that creates both product risk and product opportunity depending on where a team sits.

The BCG finding worth understanding

The article highlights a BCG study on knowledge worker performance that identified what researchers called a “jagged technological frontier.” Consultants using AI completed tasks within AI’s capabilities 25 percent faster, but were 19 percent less likely to succeed on complex tasks that fell outside what AI handles well. The implication for product managers is that AI effectiveness is not uniform across a workflow — it depends heavily on whether the specific task is one AI reliably handles or one where it tends to fail quietly.

Who it is useful for

Product managers and product leaders who want a concise signal-extraction layer over the fast-moving AI news cycle. Via writes with the assumption that readers have enough context to evaluate what the data means for their own team’s decisions. The roundup is not prescriptive, which makes it better suited to senior practitioners who want to draw their own conclusions than to those looking for step-by-step guidance.