How many creatives touch a product launch before it goes live? A photographer, a designer, maybe a video house? Tools like Dreem are collapsing that whole chain into one upload. Feed it a single product photo (even a flat lay or a phone shot) and it puts the product on AI models in about 60 seconds, across different body types, looks, and angles, plus lifestyle scenes and short video for product pages. Dreem isn't brand new, but it's the clearest example yet of where creative AI is headed: campaign-ready visuals from whatever photo you already have. For a merch pro pitching apparel, that's virtual sampling without a sample!
Marketing AI Institute makes a sharp point: Codex and Claude Code may have started as developer tools, but marketers can use them as analysts. Hand one a messy CRM export, campaign report, attribution file, or customer dataset, then give it a business question to investigate. Instead of asking AI for a quick summary, you can ask what drove conversions, why a campaign underperformed, or which segments deserve follow-up. For promo pros, the takeaway is practical: AI in marketing is moving beyond copy and images. It can turn messy sales and marketing data into direction: which accounts to prioritize, where buyers are showing intent, and what your team should do next.
Microsoft may be making the smarter long-term play in paid search for the AI era by weaving ads into Copilot, Bing, Edge, Outlook, LinkedIn, and other places where AI-assisted buying decisions now happen. One example is Audience Generation—an AI assistant that turns a plain-English customer description into targeting settings. For distributors, the takeaway is simple: buyers may ask Copilot for “good employee onboarding gifts under $50” and click only after AI shapes the shortlist. If your landing pages are vague or your feeds are messy, you may never make the cut.
How many order-status calls does your team field a week? xAI just launched a Voice Agent Builder that lets you create a live phone agent just by describing what you want in plain English. It takes care of the phone setup, adds basic guardrails, and can connect to other tools via MCP, so you can have it live on a real number in about two minutes. Point it at the calls that don't need a human — after-hours order status, basic lead qualification, simple scheduling — and keep your team on the conversations where judgment and relationships matter most. Cheap enough to test this week, give it a go!
As of July 1, Copilot Cowork is a real product with a real bill: Microsoft's promo pricing ended, Copilot Business moved from $18 to $21 per user, and the preview grace period expired. Why should a distributor care? Because Cowork is a different animal from the Copilot you've ignored since 2023. You hand it a task, it works start to finish, and it comes back with a finished deliverable: turn this quarter's client orders into a QBR deck, chase down open invoices, draft follow-ups for every quote. Under the hood, it's built on Anthropic's Claude Cowork technology (this isn't Clippy 2.0). Test it on one recurring task this month!
23% of this year's PPAI 100 runs on commonsku, so we chatted with three of them about how they got there. Sandy Gonzalez of MadeToOrder, Michael Wolaver of Magellan Promotions, and Robert Fiveash of Brand Fuel, Inc. sit down with commonsku co-founder Mark Graham to unpack the moves that unlocked their growth. Want a hint? We talked about a niche that started with a single campus, a decade-long certification most competitors wouldn't attempt, and a culture-first approach to hiring and acquisitions. Hear the highlights on skucast episode 367.
If your day-to-day workflow is built around one frontier model, you could be one policy change away from a painful outage. Take Anthropic's Fable 5: it launched June 9, was pulled June 12 over security concerns, then was only restored July 1 after the US lifted export controls. This isn't a one-off. GPT‑5.6 shipped to a small set of orgs under a government-managed access list, and "voluntary" testing is starting to look a lot like licensing. Access is also constrained by supply — Google rationed Meta's Gemini use amid a compute crunch. The practical move: standardize the "boring but critical" work on something stable like Sonnet 5, and use any limited-time access windows (like Fable 5) for the genuinely hard jobs.
🏗️ Satya Nadella says every company should build its own AI model, not just rent frontier ones.
🔧 OpenAI's Tax AI frames the pattern as harness engineering: coding agents like Codex and Claude Code doing back-office knowledge work.
📈 Exponential View says AI is scaling 3× faster than the early internet, with investment jumping from $110B to $175B and agents burning 1,200× the compute of chatbots. No wonder model access keeps getting rationed!
💰 Together AI raised $800M at an $8.3B valuation to scale open-model infrastructure, citing $1.15B+ in annual bookings as enterprises shift away from expensive frontier APIs.
🚀 AWS committed $1B to embedding small engineering teams inside customers for 45-day agentic builds, following OpenAI and Anthropic's same playbook. The NBA, NFL, and Southwest are already in. Real agent deployment is still a hands-on sport.
🖼️ Google's Nano Banana 2 Lite renders 1K-resolution product images in 4 seconds, and Omni Flash turns a still into video by voice.
🎙️ Three PPAI Top 100 distributors reveal their growth unlocks in a skucast masterclass, including how one grew 71% on a four-day workweek.