Get tomorrow's brief in your inbox
Today: DocuSign launched AI agents that cut contract review time by analyzing past negotiations automatically. Spotify partnered with Universal Music for licensed AI remixes that pay artists royalties. Google pitched an AI agent ecosystem to consumers who haven't bought into the promise yet.
DocuSign just moved from e-signature vendor to full agreement management platform with its new Intelligent Agreement Management (IAM) system, and this is a real shift for how small businesses handle contracts. The platform uses AI assistants that understand your past negotiations, company policies, and contract terms to help you analyze and redline agreements in real time. Instead of digging through old emails to remember what you negotiated last time, the AI surfaces that context automatically.
The business case is solid. Deloitte found that companies using agentic workflows with end-to-end agreement platforms see nearly 30% higher ROI compared to those that don't. For small businesses drowning in a maze of PDFs and disconnected tools, this means legal teams can actually focus on strategy instead of hunting for contract data. DocuSign partnered with Harvey and Thomson Reuters' CoCounsel to integrate with existing legal AI tools, creating a connected ecosystem that reduces friction.
The catch is integration cost and learning curve. If you're running legacy systems, implementing IAM requires upfront investment. And while automation speeds things up, you still need human oversight on complex agreements where legal nuance matters. But if contract management is a bottleneck in your business, this is worth evaluating. Read more
Zapier SDK vs. Nango for AI integrations
If you're building AI products that need to connect to other apps, you're choosing between two paths: Zapier SDK (9,000+ apps, managed auth, full ecosystem) or Nango (800+ APIs, self-host option, standalone integration layer). Zapier SDK is part of the broader Zapier platform, meaning the same actions work across Workflows, Tables, Forms, and MCP infrastructure. Nango is purpose-built for integrations only. For AI products, governance is the deciding factor: who controls app access, how credentials are protected, and whether your agents can be audited. Zapier SDK handles this out of the box. Nango gives you control but requires you to build the governance yourself. Read more
Google AI Studio: prompt to app in 10 minutes
Google AI Studio lets you type a prompt and get a working Android app on your phone 10 minutes later. One tester built three apps in an afternoon: a calorie counter and two games. The catch is they were kind of bad. The writing was terrible, bugs required fixes, and the daily token limit hits fast. But the speed is real. If you need a quick prototype or internal tool, this is viable. If you want something polished, expect to iterate and possibly pay for more tokens. Read more
Spotify Studio: AI-generated daily podcasts
Spotify launched Studio by Spotify Labs, a desktop app that generates daily briefings, podcasts, and playlists based on your listening history, email, calendar, and notes. The AI can browse the web, fetch personal info, and create a daily briefing or a podcast on demand. For example: "Create a daily audio brief for my road trip through Italy. Walk me through my day using my calendar and bookings. Recommend a dinner spot. End with a podcast recommendation." All generated podcasts save to your Spotify library. This competes with Google's NotebookLM. It's in research preview for users 18+ in 20+ markets. Read more, Read more
Datasette Agent: conversational SQL queries
Simon Willison released Datasette Agent, an AI assistant for querying your data using natural language. Add the datasette-agent-charts plugin and it generates charts too. It runs on Gemini 3.1 Flash-Lite (cheap, fast, writes SQLite queries reliably). Example: "When did Simon most recently see a pelican?" The agent wrote the SQL, ran it, and replied: "May 20, 2026." You can run it against local models. It's extensible with plugins. If you manage data in SQLite and want conversational access, this is worth trying. Read more
AdventHealth: 80% reduction in admin tasks with ChatGPT for Healthcare
AdventHealth deployed ChatGPT for Healthcare across its hospital system and reclaimed hours per week for care teams. Physician advisors reviewing utilization management cases used to spend 10 minutes per case on chart reading, criteria checking, and drafting rationales. By automating the sequence, they cut admin time by 80%. The system didn't just reduce workload. It expanded clinical capacity, sped up access to care, and improved the patient experience.
The key insight: AdventHealth treated adoption as the product, not the AI itself. They tracked messages per user per business day like any other KPI. They used peer groups (finance with finance, HR with HR) to share prompts and workflows, not centralized training. They framed AI as "time back" instead of automation, which mattered to clinicians. If you're deploying AI internally, this approach works: measure adoption, involve peers, and talk about the outcome people care about (time, capacity, stress reduction). Read more
Luma AI: From clip generation to end-to-end production agents
AI video companies learned that Hollywood doesn't want 10-second clips. It wants tools for the entire production process. Luma AI pivoted from "substitute your camera for our video model" to AI agents that guide users from concept to plotlines to characters to look and feel to final video. The agents use what they learned along the way to maintain consistency (same character appearance, same cinematic style) without complex prompting. Amazon's The Old Stories: Moses used Luma's models for LED wall backgrounds and costume rendering during shooting. If a shot looked wrong, it took one click to regenerate. This is the shift: AI for everything in the workflow, not just one step. Read more
Polyend Endless: AI-generated guitar effects for $299
Polyend launched Endless, a $299 programmable guitar pedal that pairs with Playground, an AI system that turns text prompts into functioning guitar effects. You describe the effect you want (e.g., "combination ring modulator / auto-wah"), and the AI codes it. The pedal comes with 2,000 tokens (enough for a few effects). More tokens cost $20 per 2,000. Simple fuzz might cost 20 tokens. Complex granular loopers could cost 500. The Plates gallery has 60 free effects from Polyend and the community. You can also build effects in C++ yourself. It's a well-intentioned first attempt, but iterating and testing effects takes time, and firmware has quirks. Read more
Hark raises $700M for AI personal assistant hardware
Hark raised $700M at a $6B valuation to build models and hardware for an AI personal assistant. CEO Brett Adcock (Figure.AI, Archer) launched Hark in late 2025 with $100M of his own money. The company expects to release multimodal models this summer, followed by hardware devices built for those systems. They currently have 70 employees and run a data center with Nvidia B200 GPUs. Few details are public, but the pitch is: a universal interface with the digital world, and hardware purpose-built for it. The question is whether they solve the privacy problem that wearables like Meta's glasses haven't cracked yet. Read more
Spotify and Universal Music strike AI remix deal
Spotify partnered with Universal Music Group to allow fans to create AI-generated covers and remixes of songs, with a revenue share for participating artists. The tool will be a paid add-on for Premium subscribers. Artists can opt out, but those who participate collect royalties on AI remixes. This is the first product from Spotify's "responsible AI" partnerships with major labels. The principles: artists choose whether to participate, fair compensation for those who do, and deeper fan connections. No pricing or launch date yet, but this is the licensed, artist-first approach Spotify promised as an alternative to services like Suno. Read more, Read more
Spotify audiobook creation tool powered by ElevenLabs
Spotify launched an ElevenLabs-powered audiobook creation tool within Spotify for Authors. It's invite-only beta starting June, English only. Authors can self-publish audiobooks without an exclusive contract, meaning they can publish anywhere. Spotify already partnered with ElevenLabs to accept audiobooks created on their platform, but wanted newer, more expressive voice models for authors. Spotify for Authors also expands to 10 more languages. Audiobook+ plans will allow higher listening limits, with new student and family options coming. Read more
Google pitches AI agents to consumers who aren't convinced
Google unveiled information agents (AI-powered Google Alerts), Gemini Spark (a personal AI agent for Gmail, Docs, Workspace), and Daily Brief (a digest from inbox, calendar, tasks). Most features are locked behind the $100/month Google Ultra plan. Information agents launch this summer for Pro and Ultra subscribers. Spark comes to Ultra subscribers "soon." Halo (Android notification feature) ships "later this year." The pitch is an AI agent ecosystem that works across Google products. The problem is that consumers mostly think of AI as chatbots replacing search and AI-generated slop cluttering their feeds. Google is targeting heavy users who've already bought into AI. Free users are distanced from the real-world improvements AI offers. Read more
FTC fines Cox Media Group $1M for fake "active listening" AI marketing
The FTC fined Cox Media Group and two other firms nearly $1 million for deceiving customers about an "active listening" AI-powered marketing service. The companies claimed their service listened to consumers' conversations via smart devices in real time to target ads. It didn't. Instead, they resold email lists from data brokers at a significant markup. The FTC clarified that hiding an "opt-in" in terms of service wouldn't be acceptable even if the tech worked. This is useful ammunition against the persistent conspiracy theory that mobile devices spy through microphones for ad targeting. Read more
Trump delays AI security executive order
President Trump delayed signing an executive order that would allow the government to evaluate AI models before release. He claimed he didn't like the language and didn't want to "get in the way of leading." The unofficial reason: not enough tech CEOs could make it to D.C. for the photo op. The order would have tasked agencies with developing a process to evaluate AI models for security before launch, partly in response to Anthropic's Mythos and OpenAI's GPT-5.5 Cyber (both can quickly find and exploit security vulnerabilities). One sticking point: requiring AI companies to share advanced models with the government 14 to 90 days ahead of launch. Read more
How to prompt ChatGPT for better answers
Use this skeleton: (1) Define the role ("You're a B2B marketing editor"), (2) Explain the task ("Turn these rough notes into a LinkedIn post"), (3) Give context ("Audience is ops leaders, avoid hype"), (4) Define the output ("Under 200 words, one hook, three bullets, soft CTA"). Show examples of good and bad outputs. Give only relevant information, not every thought you have. Research shows AI models don't weigh all information evenly, and performance degrades as input length grows. Be conservative with context. Read more
Zapier MCP: Seven ways to use it
Zapier MCP connects your AI tools to 9,000+ apps securely. Three use cases: (1) Start your day prepared: Pull calendar, check goals, triage Slack, surface urgent emails into a prioritized briefing with draft replies. (2) Create drafts: Generate documents flagging anomalies or trends from Google Sheets data. (3) Triage messages: Scan inbox, identify what needs attention, deliver a single brief. Zapier MCP is SOC 2 Type II certified, handles auth via OAuth, and works with any MCP-compatible AI (Claude, ChatGPT, Cursor). You decide which apps your AI can access and can revoke at any time. Read more
The shift from generative AI to agentic AI is happening fast. DocuSign, Luma, and AdventHealth are all moving from "generate a thing" to "handle the entire workflow." The economic stakes are real: BCG found agentic AI already accounts for 17% of total enterprise AI value, projected to hit 29% by 2028. But the maturity gap is dangerous. McKinsey's 2025 survey found 88% of organizations use AI in at least one function, but only 5% have capabilities deeply embedded across functions. The bottleneck isn't the technology. It's governance, ownership, infrastructure, and willingness to redesign how work gets done. Watch for tools that solve the full workflow, not just one step.