COEY Cast

GPT 5.3 Codex is being pitched as a true agentic coding upgrade that can maintain the glue work behind modern marketing systems. Hear how mid task steering, constraints, and approvals shift agents from autocomplete toys into reliable automation for content pipelines and analytics scripts. Learn why permissioning and tool connectors matter more than raw benchmarks, and how to measure real impact wi...

Long context is finally usable, but that does not mean “paste your whole Notion.” This episode breaks down StepFun AI Step 3.5 Flash and why giant context windows still fail without structure. Learn how to design a brand constitution and job packets so agents stay on voice and on policy. Get a simple way to use multi model research like Perplexity Model Council for safer claims and spicier creativ...

Amazon Ads is testing MCP as a USB C style connector for AI agents, and marketers are rethinking how much autonomy to hand over to automated workflows. Hear how MCP could run real campaign loops, where to draw the line on write access, and why approvals, audit trails, and kill switches matter. Explore Nafy AI’s all in one music generation and why brands need a sonic style guide to avoid sounding g...

Kling 3.0, Vidu Q3, and the latest in AI audio are all racing to power one prompt full ads. This episode breaks down what real coherence in AI video looks like, why cut points expose character drift, and how reference elements turn storyboards into specs. Get a practical look at Vidu Q3’s one pass video plus audio promise and why over polished sound still fails without strong scripts and pacing. T...

PrimeIntellect’s open Trinity Large, NVIDIA PersonaPlex 7B, and Gemini Agentic Vision land at once, and everything changes for automated workflows. Learn what “400B with 13B active” really means for cost, routing risk, and building your own Mixture of experts style pipelines. Explore full duplex voice for call flows, lead qualification, and creator style reads, plus the consent and disclosure rule...

Qwen3-Max-Thinking, Lucy 2.0, and CraftStory are all chasing the same prize control. This episode unpacks the Qwen3-Max-Thinking hype around Humanity’s Last Exam and explains why test-time scaling, tool use, and “open” claims often blur into configuration theater. Then it breaks down what real-time generative video like Lucy 2.0 actually unlocks for live campaigns, and why spatial precision and br...

Moltbot formerly known as Clawdbot is the latest open source automation agent obsession and it hits every nerve in the AI world. The conversation breaks down what Moltbot actually does as an orchestration and connector layer why “chat with hands” matters for real workflows and where the risk line really is once you let an agent touch inboxes calendars and automation hooks. It also unpacks the Anth...

Gemini 3.0 Ultra is promising analysis of up to one hundred hours of video per prompt, and that shift could rewrite how marketers and media teams mine their archives. This episode breaks down real workflows for webinar libraries, podcast backlogs, and brand consistency using multimodal analysis. Hunter and Riley dig into where models still fail on nuance and attribution, and why governance, receip...

Alibaba’s open source Qwen 3 TTS is pushing text to speech into a new era of local, multilingual, low latency voice pipelines with spicy zero shot cloning. Hear what this means for creators, media teams, and agencies trying to balance consent, disclosure, and scale. Then dive into Google’s Gemini powered Search AI Mode and what personalized answers do to SEO, discovery, and shared reality. Finally...

Runway Gen 4.5 makes the jump from cool demo to real workflow with better motion, character consistency, and product readable shots that actually work in ad pipelines. The conversation digs into where infinite variants become smart testing versus pure avoidance, and why creative direction needs real constraints and a motion bible to avoid same face cinematic sludge. The episode breaks down LTX Stu...

AI influencers are moving from novelty to virtual-talent factories. Higgsfield’s Influencer Studio turns static images into controlled motion avatars, then layers on Higgsfield Earn so synthetic humans can perform like media assets with payout mechanics. This episode digs into brand safety, disclosure, and how context collapse can turn polished avatars into empathy theater. Then it zooms out to Or...

Audio is turning into the control surface for modern media. This episode breaks down LTXStudio’s audio-first video workflow, from voice led editing to multi character lip sync and beat driven pacing. Hear how marketers can script sound first to test multiple visual directions without reshooting. Explore Chroma 1.0 as an open source real time speech to speech layer, what it unlocks for voice agents...

Zhipu’s GLM 4.7 Flash drops with open weights and big claims about local and edge friendly workflows. This episode breaks down what Mixture of Experts and MLA attention actually mean for VRAM, context windows, and running agents at home without becoming a full time inference engineer. Hear how agentic coding fits into real human in the loop workflows, where SWE Bench hype meets production reality,...

AI video is shifting from one-off clips to real production workflows. This episode digs into GMI Studio as a “pipeline not a demo” for longer-form cinematic video and why boring pieces like asset management, approvals, and continuity scorecards decide whether it actually works. Then it moves to FLUX klein, ultra-fast image generation, and how to avoid creative entropy when infinite variants are on...

ChatGPT Go adds a cheaper tier with GPT 5.2 Instant, higher limits, image gen and file uploads while OpenAI quietly tests sponsored answers inside chat. Learn what assistant native ads might look like, how to keep answers sacred, and how marketers can build structured truth packs instead of spam. Then dive into Alibaba’s Qwen3 VL Embedding and Reranker for multimodal retrieval across screenshots,...

Claude 3.5 Sonnet, SAM Audio, and Step-Audio-R1.1 just changed the audio and agent game at the same time. This episode breaks down why “insanely fast” models expose weak links in your stack, how to keep agent workflows from melting your budget, and where humans still own taste and accountability. Get into promptable audio surgery with SAM Audio, local audio reasoning with Step-Audio-R1.1 open weig...

GLM-Image from Zhipu is an open source image model built to handle posters, slides, and infographics where text accuracy actually matters. Hunter and Riley break down how its hybrid architecture aims to fix cursed typography, what to stress test before trusting it with paid campaigns, and how to slot it into a prompt to poster pipeline that still ends in Figma. They also dig into Yuan 3.0 Flash an...

Meituan drops its open source LongCat-Flash-Thinking-2601 reasoning agent and the internet yells “thinking is solved.” This episode breaks down what that actually means for creators and marketing teams, from Heavy Thinking Mode to million token context and why you still need retrieval and guardrails. Then it is OpenAI’s GPT 5.2 Codex sliding into the Responses API and what that unlocks for safe co...

Google DeepMind’s Genie 3, Kyutai’s Pocket TTS, and Sakana’s DroPE are reshaping how content gets made and deployed. Hear what world models mean for interactive marketing experiences that behave more like games than landing pages, including guardrails and security risks. Learn how five second voice cloning on CPU rewires voiceover workflows, A B testing, and brand trust. Get a plain English breakd...

Anthropic is rolling out Claude Health and Claude Cowork and kicking off the era of vertical AI agents. This episode breaks down what “privacy first” actually needs to mean for healthcare workflows, where open source fits without becoming a rogue diagnosis engine, and why insurance and messy medical records are the first things to break. Then it shifts to office agents, Cowork style products, and...

OLMo 3, Nemotron Speech, and Fun-Audio-Chat are pushing three very different ideas of what “open” and “real-time” actually mean for creators. This episode breaks down how end-to-end transparency, dataset lineage, and reasoning traces affect trust and control for teams building with open models. It explains where preference optimization helps brand voice and where it turns into pure vibes. Then it...

Anthropic’s next generation constitutional classifiers promise fewer jailbreaks and fewer bogus refusals by layering interpretability probes with smarter safety filters. Claude Code 2.1 pushes deeper into agent style background tasks inside developer and operator workflows, along with the reliability risks that come with autonomous CLIs. OpenAI’s upgraded speech stack improves transcription and sp...

CES 2026 put AI into everything from your TV to your earbuds. This episode breaks down Google Veo 3 on Google TV and what couch based text and voice to video really means for brand teams and creators. It covers open source friendly video stacks for teams that want control instead of closed ecosystems. It looks at StreamUnlimited Voice LLM integrations and what it takes to ship safe branded assista...

CES 2026 puts agentic AI everywhere from Boston Dynamics Atlas doing Gemini powered physical tasks to Lenovo Qira living across your devices to IAS Agent reshaping media ops. This episode breaks down what is real autonomy versus cinematic choreography, how to spot PR smoke and mirrors, and where open source robotics gets risky. Hear how on device personal agents will collide with workplace governa...

CES 2026 is all about AI leaving the cloud and moving into your stuff. This episode breaks down NVIDIA’s Alpamayo for autonomous driving and LTX-2 for local video generation on RTX PCs, and what “open-source” actually means when the stack is hardware-shaped. Hear how explainability, synthetic edge-case training, and model governance map directly to content ops and brand safety. Learn why local-fir...