PLUS: Adobe AI drives 693% shopping surge, DoorDash safetymishaps drop 50%

"Life is much less a competitive struggle for survival than a triumph of cooperation and creativity.” - Fritjof Capra

Afternoon All, Corporate legal frameworks are fracturing under pressure from new agentic AI deployments - with Amazon's lawsuit against Perplexity setting the stage for high-stakes battles over autonomous systems.

As enterprises adopt specialised AI networks for efficiency gains, this landmark case forces us to ask: When an AI team makes decisions spanning multiple vendors, where does liability actually land?

Today's dots:

• Amazon sues Perplexity over agent shopping AI
• AI-powered referrals drive 693% shopping surge
• DoorDash halves safety incidents with layered AI
• Corporate legal risks of agentic AI networks
• The growing AI productivity paradox

Agentic AI Teams Rewrite Corporate Playbooks

Here's the thing: Enterprises are ditching single-AI solutions for networks of specialised agents. Systems that boost efficiency but create novel legal risks, as highlighted by Amazon's ongoing lawsuit against Perplexity's shopping AI.

Let's unpack that:

• Google's agent framework envisions self-orchestrating AI teams that dynamically assign tasks - like human departments collaborating in real-time
• Companies report 20-35% efficiency jumps letting specialised agents handle siloed workflows (procurement, compliance etc.) while maintaining audit trails
• Amazon sued Perplexity after its Comet AI disguised automated purchases as human activity - a landmark case testing if bots must disclose their identity during transactions (Perplexity's response)
• Corporate lawyers warn unchecked agent networks could degenerate into liability mazes – with actions spanning vendors making blame attribution murky
• Early adopters in healthcare and finance already cut contract review times by 60% using agent teams chaining diagnostic, negotiation, and compliance AIs

If you remember nothing else: Autonomous AI squads are inevitable workplace collaborators, but their actions need guardrails before they act. Businesses building governance frameworks today will outpace rivals drowning in agent-enabled complexity tomorrow.

AI drives 693% shopping surge… but not everyone's buying in

Here's the thing: AI-powered shopping referrals exploded 693% YoY in 2025's holiday season (Adobe data), as Amazon sued Perplexity over its Comet AI agent in a landmark e-commerce dispute - revealing both AI's retail potential and emerging tensions.

Let's unpack that:

• AI now converts 31% better than traditional traffic (Adobe study), with AI-referred shoppers spending 45% longer browsing - making them prime targets for DTC brands
• Urban shoppers are rapidly adopting AI helpers (48% usage rate) versus rural consumers (27%), creating uneven playing fields for nationwide campaigns
• Thanksgiving saw AI conversions spike 54% thanks to last-minute deal hunters, proving urgency turbocharges agent-assisted shopping
• Amazon claims Perplexity's AI breaks ToS by logging in as users, prompting a lawsuit painting Comet as digital "trespass" (official statement)
• Perplexity calls this "bullying," positioning AI agents as personal assistants consumers can ethically use (their rebuttal)

If you remember nothing else: AI referral traffic isn’t coming back to traditional search - shoppers clearly want agentic help. But its wild-west phase is ending: expect platforms to fight for control like Amazon did, making compliance your next competitive edge.

DoorDash's AI Safety Net Catches 50% More Trouble

Here's the thing: DoorDash just slashed safety incidents by half using a clever layered AI system called SafeChat, proving you can scale safety without slowing down delivery ops.

Let's unpack that:

• Their three-step text filter auto-clears 90% of messages instantly using a low-cost API, then applies precision-tuned LLMs to catch trickier cases - cutting moderation costs while maintaining 99.8% accuracy.
• Phase 2's smarter architecture processes 99.8% of traffic in under 300ms, only flagging 0.2% for slower (3-second) deep analysis - perfect for time-sensitive deliveries.
• Image moderation handles hundreds of thousands of daily uploads with computer vision models fine-tuned through human feedback loops - reducing false positives that could disrupt Dasher workflows.
• Voice call monitoring started in 'observe-only' mode - a smart sandbox approach letting them validate thresholds before enabling real-time interruptions in sensitive situations.
• As their engineers noted in a LinkedIn announcement, this isn't just tech: 'SafeChat builds confidence through AI innovation and thoughtful design' protecting both workers and customers.

If you remember nothing else: This proves safety tech doesn't have to mean slower service. DoorDash's phased, human-in-the-loop approach sets a template for any platform handling high-volume interactions - food delivery today, maybe your industry tomorrow.

Amazon vs Perplexity: AI Agents Spark Legal Showdown

Here's the thing: Amazon is suing Perplexity over its agentic shopping assistant Comet, claiming the AI violates Amazon's Terms of Service by disguising automated browsing as human activity - revealing how existing tech legal frameworks are straining under new AI capabilities.

Let's unpack that:

• Amazon alleges Comet engages in 'computer fraud' by bypassing bot detection while completing purchases, effectively creating an undetectable third-party layer between shoppers and Amazon's platform
• The retail giant filed a federal lawsuit days after Perplexity ignored cease-and-desist demands, claiming the AI introduces security risks and sabotages Amazon's recommendation algorithms
• Perplexity's public rebuttal blog post argues users have the right to AI-powered shopping assistants, comparing Comet to employing a human personal shopper
• This case could set precedent for how platforms handle 'agentic browsing' - where AI tools act autonomously using human credentials

If you remember nothing else: This legal battle highlights how quickly AI innovations are outpacing existing digital commerce frameworks. Platforms will likely revamp ToS to explicitly restrict or regulate agentic AI interactions in the next 12 months.

The AI Productivity Paradox: Why Tools Aren't Translating to Time

Here's the thing: 40% of workers say AI saves them no time despite executives claiming double-digit hourly savings each week - with MIT-affiliated research showing developers actually take 19% longer with AI assistance (Section study).

Let's unpack that:

• The productivity gap is stark: developers thought AI made them 24% faster but were paradoxically 19% slower in controlled trials - likely due to debugging and refining AI outputs (METR research)
• 85% of employees who initially report AI time savings later admit wasting that reclaimed time correcting errors, per Wall Street Journal data
• Microsoft CEO Satya Nadella flagged this disconnect at Davos, warning AI must show real-world efficiency gains beyond boardroom hype
• Retail workers report the lowest AI productivity boosts while tech sees patchy benefits - with 85% of workers still considered AI beginners

If you remember nothing else: Successful AI adoption needs better implementation planning than most companies currently invest. Tools alone won't boost productivity - trained workflows and realistic expectations do.

The Shortlist

AI Agents may hit computational limits during complex tasks according to new mathematical proofs from Sikka Research. The study suggests agentic AI could struggle with multi-step logic beyond certain complexity thresholds.

Chatbots reshape daily routines as Replika users report AI acting as 'practical companions' during chores and isolation. The Loughborough University study found most users maintain awareness of AI's artificial nature despite emotional connections.

Apple betting on wearable AI pins and glasses despite past failures. Scientific American reports both companies aim to launch devices by 2027 that interpret surroundings through cameras and microphones.