Your AI Agent Might Work Perfectly. That Could Be the Problem.

PROJUKTI BRIEF

1  A former DeepMind lead builds an agent that hands off to real people

Shivani Poddar, who led engineering at Google DeepMind, has launched Fo through her startup Wajo. It is a personal agent with a safety net: when a task is beyond what AI can finish alone, Fo brings in a human assistant to complete it. Wajo says Fo beats rivals on trust and task completion, though that benchmark has not been independently checked. See Fo in action.

2  Anthropic ships a faster, cheaper Sonnet while ElevenLabs gives its voices more feeling

Claude Sonnet 5.5 is 30% faster and 30% cheaper than Sonnet 5, and it edges out Fable 5.1 on one performance index. Anthropic released a prompting guide alongside it. Separately, ElevenLabs launched Eleven v4, a new architecture built to handle tone, pacing, and emotion more naturally than any of its earlier voice models.

3  SpaceX AI lets teams build role-based agents and share them

With Team Bots, teams on Teams and Enterprise plans can design agents around a specific job or workflow and pass them around the office. Each bot keeps personal conversations separate from the skills the whole team shares. Grok Bot also picked up a Finance integration, so it can now help manage your money.

B2B customer issues move across teams and systems, not through a single chat window. A new Harvard Business Review Analytic Services briefing paper, sponsored by Front, breaks down where AI tools fall short in B2B service and what to ask before you invest.

Get the Free Report

PROJUKTI FORWARD

The flip side. Most worries about AI agents focus on them going off the rails. A quieter problem is showing up instead. Agents that follow instructions perfectly, negotiating bills, canceling subscriptions, and booking tables, are putting pressure on systems that were built for the pace of human requests.

What it looks like. One agent asked to land a dinner reservation hit Resy hundreds of times an hour, around the clock. Companies are also seeing customer service lines flooded by agent calls that their staff and phone systems were never sized to handle.

Where it could get serious. Apollo chief economist Torsten Slok raises a bigger scenario. If millions of households tell their agents to find the best savings rate, money could move out of lower-paying banks and into the top payer all at once. That is how a bank run starts, except this one would be triggered by software acting in good faith.

What to watch. Each request makes sense on its own. The trouble comes from the total. As agents spread, restaurants, call centers, and banks will need rules for automated traffic, and the open question is whether those rules arrive before the next pileup.

PROJUKTI SNAPSHOT

🧠 Four Words, Better Answers. A Modal engineer shared a four-word prompt he says makes chatbots feel noticeably smarter (4K bookmarks).

📝 When to Pick Sonnet. Anthropic's new build guide for Sonnet 5.5 explains which jobs suit Sonnet and which still call for Opus.

✌ Meme Parade. An AI video of a woman strolling past a street full of famous meme characters has 2M likes on Instagram.

👀 Big Hire. MongoDB's CEO is leaving to run enterprise AI at Meta, and Zuckerberg's announcement has 3.5M views.

😅 Opening the Books. Anthropic has filed to go public, and the financials in the filing are drawing a lot of raised eyebrows.

📈 Thesis Investing. A new startup turns your market hunches into stock portfolios you can invest in, using AI to pick the holdings.

PROJUKTI PROMPT'S