The AI Marketing Podcast

Will you let an AI agent act on your data with approvals and an audit trail—or keep it as a brainstormer? Microsoft Copilot Cowork in 365 uses Claude to run tasks across Outlook, Teams, OneDrive, SharePoint, and Excel, with governance baked in; Frontier and E7 previews. Google Gemini Embedding 2 maps text, images, video, audio, and PDFs into one searchable representation, with cross-file drafting in Docs and Sheets. The takeaway: governance and execution determine AI value, not speed.

Are your AI vendors a supply-chain risk your board will demand you fix this quarter? The DoD designates Anthropic a supply-chain risk, OpenAI's Pentagon deal is under scrutiny, and you'll see defense contractors pulling back. GPT-5.4 ships with models that use your computer, capable of taking actions across apps; Google Workspace CLI lets you control Gmail, Drive, Calendar, Docs, and Sheets with mandatory access audits. Your board will fund AI on control and continuity, not scores.

Are you ready to let AI click through your desktop apps and run end-to-end workflows for you? GPT-5.4 brings native desktop control, on-demand tool loading, and near-human OS task performance with token savings; pricing is shifting from seats to outcomes in SaaS, so stress-test the impact of a modest seat reduction on your forecast. Walk away with a judgment on mapping allowed AI actions and human review before the CFO sees the line item.

Are you ready to audit payoffs, not tools, as desktop agents take over? OpenAI's GPT-5.4 unlocks desktop autonomy: agents click, type, and complete 75% of end-to-end tasks, cutting tool overhead in complex workflows by about 50%. Boards debate misstep liability, and your procurement pushes to embed top vendors like OpenAI, Anthropic, and Google. You’ll walk away with a judgment that value moves from seats to outcomes, demanding auditable proof and clear ownership for failures.

Are you betting your AI budget on seat licenses or on orchestration that actually moves work? GPT-5.4’s Thinking mode plans and operates your desktop, uses Tool Search to cut token waste, and nears human accuracy on multi-step tasks. Pricing shifts from seat licenses to orchestration; map vendors as orchestration, memory, raw model, or seat tax, and push one renewal toward an outcome-based pilot. You’ll walk away seeing who actually controls action in your stack before Finance does.

Your biggest decision this week isn’t which feature to ship—it's whether the default AI layer will save you time or cost you rework. Claude hit #1 in the US App Store amid a government-terms dispute, nudging you to draft a one-page AI vendor narrative your board would accept. Google Gemini 3.1 Flash-Lite is a cheaper, high-volume model for small tasks; test taxonomy cleanup and escalate only if it misses. In 2026, winners prove reliability, governance, and cost control at scale.

Are you still paying for APIs, or can you self-host and govern a model? Alibaba's Qwen 3.5 9B is open-weight under Apache 2.0, self-hostable on 6–12 GB VRAM when quantized, with throughput on vLLM. OpenAI's GPT-5.3 Instant raises procurement and compliance questions for regulated data, so map which models touch sensitive info and plan switch-ready fallbacks. You’ll walk away seeing open-weight plus context as a governance decision—not just a cost, an asset you can audit and scale.

Are you betting on AI that can actually operate your tools, not just draft copy? Alibaba’s Qwen 3.59B open-weight model runs on a laptop, solving data residency and lower inference costs. LTX 2.3 brings high-quality AI video to the desktop; run a 30-minute pilot to render a 15-second ad and compare time and cost to your current pipeline. You’ll walk away with a governance lens—costs, risks, and who owns the blast radius.

Are you still paying for APIs, or can you self-host and govern a model? Alibaba's Qwen 3.5 9B is open-weight under Apache 2.0, self-hostable on 6–12 GB VRAM when quantized, with throughput on vLLM. OpenAI's GPT-5.3 Instant raises procurement and compliance questions for regulated data, so map which models touch sensitive info and plan switch-ready fallbacks. You’ll walk away seeing open-weight plus context as a governance decision—not just a cost, an asset you can audit and scale.

Are you still paying for APIs, or can you self-host and govern a model? Alibaba's Qwen 3.5 9B is open-weight under Apache 2.0, self-hostable on 6–12 GB VRAM when quantized, with throughput on vLLM. OpenAI's GPT-5.3 Instant raises procurement and compliance questions for regulated data, so map which models touch sensitive info and plan switch-ready fallbacks. You’ll walk away seeing open-weight plus context as a governance decision—not just a cost, an asset you can audit and scale.