A Week When Copilot Entered Slack and Teams Conversations, and Sign Language AI and Game Agents Appeared Side by Side

From August 17 to 23, the center of official materials was the public preview putting GitHub Copilot into Slack and Microsoft Teams. Google DeepMind released a sign language-to-text model for deaf and hard-of-hearing users and a general-purpose game agent aimed at persistent worlds, and OpenAI opened a blog on long-term social change while disclosing a ChatGPT Work use case.
The Week's Flow
Official materials published from August 17 to 23, 2026, put more weight on everyday conversation windows and attaching agents to actual users' hands than on scorecards. Weekly keywords also overlap on GitHub, Copilot, work, experience, Slack, agent, sharing, Microsoft, Teams, sign language, and user. These were not isolated research announcements; a single line connected the move to slot agents into messengers people already use, accessibility features, game worlds, and release work.
The densest axis was GitHub. Slack's GitHub integration brings the agent capabilities of GitHub Copilot CLI and the GitHub Copilot app into Slack as a public preview, and in Microsoft Teams a shared agent session begins—visible to everyone so they can align on direction—by mentioning @GitHub in channels, threads, and direct messages. This was a week when an assistive tool that had stayed in the code window stepped up as a participant in team conversation. In the same period, Google DeepMind introduced SL2T, a sign language-to-text model that drives new sign language features for deaf and hard-of-hearing users, and said that, drawing on 15 years of game AI research from Atari to EVE Online, it would join hands with game developers to put general-purpose agents like SIMA 2 into persistent worlds. OpenAI opened the blog AI Futures, which addresses changes transformative AI could bring to power, governance, the economy, and personal freedom, and disclosed a case in which Stampli, amid fixed deadlines and design resources tied up elsewhere, used Codex and ChatGPT Work to compress weeks' worth of release production into a matter of days and cut release time by 68%.
Operational details such as which organizations the preview covers, pricing, the product screens that get sign language features, and the list of game partners are not confirmed in currently published official materials. This week was less a week that proved feature completeness with numbers than a week that wrote down, on official channels, the places agents would enter.
Weekly Hot Issues
Topic: GitHub Copilot opens as shared agent work in Slack and Teams
GitHub said Slack integration has opened a public preview that brings the agent capabilities of Copilot CLI and the Copilot app into Slack. Users can work with @GitHub. In the same flow, Microsoft Teams turns conversation into a collaborative agent session everyone can see and steer together, starting GitHub Copilot work by mentioning @GitHub in channels, threads, and direct messages. The two announcements are closer to overlaying agent sessions on Slack and Teams threads that are already on than to asking anyone to install a new tool. Which workspaces the public preview opens to, and how permissions and audit logs are tied, are not confirmed in currently published official materials. Still, with the week's keywords clustered on GitHub, Copilot, Slack, Teams, sharing, and agent, the practical issue this week is not model replacement but joint instruction authority inside the conversation channel.
Topic: DeepMind backs features for hard-of-hearing users with a sign language-to-text model
Google DeepMind introduced SL2T, a breakthrough sign language-to-text model that drives new sign language features for deaf and hard-of-hearing users. The direction the title points to is putting a feature in users' hands, not a research paper. On which app screens, which sign systems, and which devices it turns on directly is not confirmed in currently published official materials. Even so, the reason sign language and user rose together among the week's keywords is clear: in the same week agents entered developer messengers, official channels put out, side by side, a converter that links users who do not hear to text. The backbone of this announcement is not a story that bolts on accessibility as an extra, but that it named a model converting sign input to text as the engine of the feature.
Topic: Carrying 15 years of game research, SIMA 2 is to be put into persistent worlds
Google DeepMind said that, on the back of 15 years of game AI research from Atari to EVE Online, it will partner with game developers to build general-purpose agents like SIMA 2 and open play experiences across persistent worlds. It is a declaration to have general-purpose agents move inside worlds that continue without a break, beyond short arcade tasks. Which titles attach when, and SIMA 2's control scope and evaluation methods, are not confirmed in currently published official materials. Just as shared sessions in Slack and Teams turn human conversation into agent work, this side named long-lived spaces like game servers as the agents' stage. The sign language model from the same company is a different topic, so they should not be mixed into one lump. One thread is accessibility; the other axis is persistent game worlds.
Topic: OpenAI opened a blog that asks about long-term change and disclosed a case of shorter release work
OpenAI introduced a new blog, AI Futures, on how transformative AI could change power, governance, the economy, and personal freedom. It is clearly a writing space that puts institutions, the economy, and freedom in front, in a grain different from product release notes. Separately, it disclosed the Stampli case. With deadlines fixed and design resources tied to other work, Stampli used Codex and ChatGPT Work to compress weeks' worth of release production into a matter of days, and it is written that release time fell 68%. The word work can look as if it overlaps GitHub's shared agent work and OpenAI's ChatGPT Work, but the former is a joint session inside a messenger and the latter is a work-tool case that compressed release production. From which definition of time and which comparison baseline the 68% came, down to the detailed formula beyond the title and summary, is not confirmed in currently published official materials.
Next Watch Points
Topic: Do the Slack and Teams public previews settle as joint instruction in real threads?
The next point to watch is not a teaser for a new messenger integration. It is whether the Slack public preview already open and Teams' shared agent sessions settle into work where one @GitHub mention lets the whole channel see and correct direction. Which organizations and permission models the preview opens to, and who can take the code and issues that appeared in the conversation, are not confirmed in currently published official materials. The watch is whether, more than the number of integrations, official cases keep showing a scene in which several people push the same agent session together inside one thread.
Topic: Do the screens and language scope attached to the sign language-to-text model come into view?
SL2T was introduced as driving new sign language features, but on which product surfaces it turns on, and the scope of supported sign languages and devices, are not confirmed in currently published official materials. Watch whether the direction of putting it in users' hands becomes a switch on an actual settings screen, beyond a research demo. The same applies on the game side. The next check is whether SIMA 2 and persistent-world partnerships become concrete under specific developer and title names. For both branches, watch not the model name being repeated but whether names attach to the screens hands actually reach and to the servers where play continues.
Topic: Do further AI Futures posts and a follow-up to the ChatGPT Work case appear?
AI Futures was introduced as a writing space that explores power, governance, the economy, and personal freedom. The next watch is whether it ends as a single declaration or whether follow-up posts continue in the same frame. For ChatGPT Work, one official-channel case has gone up: Stampli's 68% cut in release time and compression from weeks to days. Whether the same tool is written with a similar definition of time on other release work, or remains a one-off story, is how to gauge whether Work is a work path beyond a product name. Comparison formulas and application conditions are not yet open outside official materials.
Sources
- GitHub Changelog, The New GitHub Copilot Experience in Slack: https://github.blog/changelog/2026-08-21-the-new-github-copilot-experience-in-slack
- GitHub Changelog, Shared Agent Work with GitHub Copilot in Microsoft Teams: https://github.blog/changelog/2026-08-21-shared-agentic-work-with-github-copilot-in-microsoft-teams
- Google DeepMind, Putting Sign Language AI into Users' Hands: https://deepmind.google/blog/putting-sign-language-ai-into-users-hands/
- Google DeepMind, 15 Years of Game AI Research from Atari to EVE Online: https://deepmind.google/blog/from-atari-to-eve-online-building-on-15-years-of-ai-research-in-games/
- OpenAI, Introducing AI Futures: https://openai.com/index/introducing-ai-futures
- OpenAI, Stampli Case of Cutting Release Time 68% with ChatGPT Work: https://openai.com/index/stampli
This week's official materials gather around attaching agents to existing conversation channels such as Slack and Teams, to sign language users, to persistent game worlds, and to release work, rather than around new model scores. Public previews and the case figures do not explain operating conditions, but the fact that the places agents enter are tools already in use, not the lab, changes the practical watch points.