What it means for us

'World-first' vaccine designed by artificial intelligence

Not sure how this one didn't make top headlines. The BBC reported on Friday that AI has been used to develop a "fundamentally new" type of vaccine that could protect against large swathes of viruses and prevent pandemics — with the team already developing separate vaccines that could tackle flu and Ebola.

So, look: it may yet destroy the universe, the multiverse and all life throughout time (unless the anti-AI crew is wrong) — but this is good news.

OpenAI reportedly wants ChatGPT to become a work "superapp"

Reuters, citing the Financial Times, reports OpenAI is planning a major ChatGPT overhaul to make it more like a multifunctional work platform — giving more prominence to Codex, AI agents, image generation, and partner services including Canva and Booking.com. OpenAI didn't immediately confirm, so treat this as reported plans, not delivered product. But the direction is unmistakable.

This echoes something we've discussed in our workshops — the slow, steady collapse of the distinction between "AI tool" and "work surface." Stop asking "which tool do I use?" and start asking "who controls the layer where my work happens?"

Satya Nadella says AI agents need identities, permissions, and audits

Microsoft's CEO said AI agents should be managed more like employees: with identities, sandboxes, permissions, policies and auditability. If agents are going to act on behalf of the organization, they need what every trusted colleague needs — a clear role, clear permissions, and clear accountability.

The relevance to communications is obvious to anyone who's ever mopped up a PR disaster. If you'd rather not clean up after your firm's chatbot fires off an unsolicited email blast, this one is worth paying attention to.

Microsoft pushes agents into Teams and Microsoft 365 Copilot

Agents can now be published directly into Microsoft 365 Copilot and Teams, with "autopilot agents" in public preview. These agents can have their own identity, email, calendar, OneDrive and Teams access — and a place in the org chart. Microsoft's framing has shifted from "Prompt → Response" to "Goal → Ongoing execution → Checkpoints → Collaboration."

Organizations will soon need language for what agents are authorized to do, how they escalate, and why employees should trust them. That language doesn't write itself — guess who gets asked to do that job?

Meta launches Business Agent across WhatsApp, Messenger and Instagram

Meta unveiled an AI business agent handling customer questions, lead qualification, appointment booking, sales and escalation to human staff. More than a million businesses had already used earlier chatbot versions.

The comms question here isn't technical — it's about brand voice. Who writes the agent's tone? Who approves its limits? How quickly can a human step in when the brand relationship gets delicate? These are communications decisions, and right now most organizations are treating them as IT decisions. The more things change…

Microsoft frames enterprise AI as a governed operating system, not a collection of pilots

At Build 2026, Microsoft introduced "a new category of agents called Autopilots. Autopilots are always-on agents that work autonomously, with their own identity, and act on your behalf." Not sure about you, but "autopilot" doesn't leave me with the warm and fuzzies. Sounds a bit like we're going to let it fly on its own. In any case, Microsoft is introducing Scout as its first autopilot agent.

Here's what TechCrunch had to say: "The system also comes with extensive security protections, meant to address concerns of unsupervised AI agents running amok, a real issue that OpenClaw surfaced earlier this year when one agent was reported to have acted erratically inside a researcher's inbox (among other examples). Scout will come with a built-in 'policy conformance system' that will continuously check whether the system is operating according to set guidelines, and each conformance check will produce its own audit trail."

The research-synthesis tool most speechwriters aren't using

NotebookLM is a source-grounded AI research assistant — you upload your documents (reports, transcripts, briefing papers, previous speeches) and it answers questions, generates summaries and builds outputs from those specific sources rather than general training data. The 2026 updates added Audio Overviews (a podcast-style dialogue between two AI hosts discussing your sources — not weird at all), Video Overviews, and Reports.

For speechwriters facing big events, the workflow is immediate and a lot "safer" than a general search: upload the agenda, your background reports, a few previous speeches and a briefing note, then ask it to identify the core tensions, suggest the strongest argument, and draft a three-point outline — all grounded in what you uploaded. Less hallucination risk. Genuinely underused. Just don't ask it to write the damn speech.

Private AI speech coaching for high-stakes delivery

Yoodli is an AI speech-coaching platform that runs during live Zoom, Google Meet or Teams calls, delivering real-time private prompts as you speak — slow down, cut the filler, manage pace. It also supports solo practice with AI-generated personas, and tracks filler words, pace, eye contact and word choice against "professional benchmarks," including TED presenters.