OpenAI has begun testing Sponsored Agents in ChatGPT, allowing US advertisers to turn a clicked ad into a labelled, persistent conversation with a brand's agent. The move, announced September 16, shifts the creative unit from a static impression to a sequence of questions that can shape a buyer's understanding before they ever reach the advertiser's site. Google followed with production-grade voice benchmarks for Gemini 3.8 Live, Adobe confirmed its fourth annual Firefly training-data bonus now includes Stock video, and an IBC award went to a broadcaster-led framework for AI assistant directors in live production. The common thread across all four signals is that AI is now close enough to commercial decisions that teams need to define whose facts, rights and authority travel with it.
Ads become conversations with defined - and undefined - limits
OpenAI's Sponsored Agents launch includes natural-language campaign creation through an Ads Manager plugin, suggested copy and imagery, optional contextual text customisation, and integrations with HubSpot and Shopify. The test is limited to selected US advertisers. The format gives a brand room to explain a complex product, but it also turns product data, exclusions, availability, claims and escalation behaviour into advertising material. OpenAI's announcement describes the format and controls; it does not publish independent conversion, accuracy or brand-safety results.
For creative and marketing teams, the practical step is to write an approved-answer set before writing the agent's personality. Connect only current product, price, territory and policy data. Make uncertainty and human escalation explicit, block unsupported comparisons, and log the question, retrieved evidence, answer, disclosure, click-out and downstream outcome. In a pilot, compare qualified visits and assisted conversion with a conventional ad while separately scoring factual errors, unsafe claims, abandonment and human handoffs.
Voice models get inspectable - but not yet production-ready - scores
Google introduced Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking on September 15, then updated the announcement two days later. The lower-latency model supports visual grounding, automatic switching across 97 languages and asynchronous tool calls while dialogue continues. Extended Thinking targets more complex multi-step work. Google also launched Gemini 3.5 Transcribe for 85-plus languages, files up to one hour, timestamps and speaker diarisation, with SynthID watermarking on generated audio.
The launch stands out because Google published several evaluation conditions rather than relying only on a polished voice demo. It reports an 82.6 Speech-to-Speech Quality Index score, 68.6% on tau-Voice, 35.1% on Sierra's banking version of that benchmark and 97.7% on Big Bench Audio. Those numbers make the launch more inspectable, but they still combine external leaderboards with supplier-reported tests and do not predict performance in a noisy studio, multilingual production office or customer call. Teams evaluating voice agents should build a 30-call evaluation set from the actual accents, interruptions, room noise, visual references and tool actions their workflow contains, and store the audio, transcript, tool trace and reviewer score together.
Training-data payments become recurring - and video enters the calculation
Adobe updated its Firefly FAQ on September 16 ahead of the September 17 contributor payment. Eligible Adobe Stock contributors whose photos, vectors, illustrations, videos or generative-AI content were considered for Firefly training receive a discretionary bonus. The 2026 calculation uses content considered between June 3, 2025 and June 2, 2026 together with the licences those assets generated in that period. Adobe confirmed this is the fourth year of the programme and that Stock video is now used in Firefly model training.
Recurring payment makes training-data compensation operational rather than hypothetical, and the addition of video brings filmmakers' motion assets into the same economics. The limits matter just as much. Adobe does not publish each asset's contribution, the formula or future payment commitment, and contributors cannot opt out while remaining under the existing Stock terms. A bonus demonstrates a compensation mechanism; it does not by itself explain consent, attribution, model influence or whether the amount tracks the value created. Agencies buying generative services should ask vendors how creator payment, removal requests and future model versions are handled - not settle for the phrase "commercially safe."
Broadcasters back an open agent control layer for live production
At IBC2026, the 2025 Accelerator Project of the Year award went to AI Agent Assistants for Live Production. The project demonstrated an AI Assistant Director orchestrating specialist production agents through voice and natural-language control. ITN, BBC and Channel 4 championed the work with vendors including CUEZ, Shure, EVS, Moments Lab and Google Cloud. IBC highlighted an open, vendor-agnostic framework and the preservation of editorial control.
Live production is where agent claims meet unforgiving timing. A system that can coordinate cues, microphones, media, highlights or graphics across vendors is more consequential than a chat interface, because a plausible mistake can go to air. The broadcaster-led emphasis on an open control layer suggests that interoperability, traceability and human override are becoming part of the product requirement. The award recognises a proof-of-concept shown in 2025; it is not evidence of unattended broadcast deployment or a published reliability benchmark. Production teams should run the assistant director in shadow mode before granting control, map every natural-language instruction to a typed command with a named human owner, and require a hard kill switch with deterministic fallback rundowns.
Why this matters for creatives, marketers and writers
The commercial surface of AI is widening fast. A sponsored agent can shape a buyer's understanding after an ad. A live voice model can act while a conversation continues. A training-data policy can turn a creator's catalogue into model input. An assistant director agent can coordinate systems inside a control room. For the people who write the copy, produce the assets, manage the brand voice or sign off on the claims, the question is no longer whether the AI can produce fluent output. It is whether the system is authorised to represent or change something - and whether the team can trace every automated decision back to an accountable source, owner and route to safety. Write the authority boundary beside the capability test, and test the edge, not the demo.
Your membership also unlocks:
