Realtime AI News
Adobe ships Photoshop Relight on Mac, but agent MCP support slips to 2027
Adobe has released Photoshop's Relight feature for Mac, according to The Mac Observer, giving users a new way to change the lighting in an image. The same report says support for AI agents through MCP will not arrive until 2027.
Adobe's latest Photoshop update carries two messages on two different schedules. According to The Mac Observer, the Relight feature is available on Mac now, while support for AI agents built on MCP is not expected until 2027.
The split is concrete: image-side tooling ships first, agent-side interoperability comes later. Designers and retouchers get the new lighting tool immediately; developers and teams who want external agents to drive Photoshop have more than a year to wait.
Relight addresses a well-defined task in photo post-production: changing where the light in a shot appears to come from, work that has usually meant manual masking and curves. Productising that step continues Adobe's pattern of embedding AI into workflows people already use rather than asking them to adopt new ones.
MCP sits at a different layer. It is the protocol direction for connecting external AI agents to software, and support would let an agent act inside Photoshop rather than merely produce an image beside it. Putting that on a 2027 timeline says Adobe is moving deliberately on handing over control.
The gap between the two dates is the signal. Generative features change what gets made; agent access changes who does the making. The first is comparatively contained, the second raises questions of permission, reversibility and whether a workflow stays predictable, which is where software vendors tend to slow down.
The report does not specify system requirements, which subscription tiers include Relight, or how much it costs, and it does not say when in 2027 MCP support lands. What is confirmed is the platform and the window: Mac today, agents later.
For the creative software industry the cadence is a useful reference point. The past two years have been about stacking generative features; the agent integration that would actually reshape workflows is arriving visibly more slowly.
What to watch: whether Windows follows, whether Relight lands in every subscription tier, and whether 2027 brings MCP support to Photoshop alone or across Adobe's other creative applications.
Why it matters
The two-speed rollout shows creative software is still competing on features while the agent integration that would reshape production remains closed. For teams built around Photoshop, the near-term gain is an efficiency tool; the workflow change is at least a year out.
Nearby Updates
All11/11, 21:02
Anthropic Sets $1.4T IPO Valuation Target in S-1 Filing
Anthropic has set a $1.4 trillion valuation target in its S-1 filing, a figure Reuters, The New York Times and The Guardian reported on the same day. If the listing prices near that level, it would be the first time the AI build-out is valued by public markets rather than private rounds.
11/11, 20:53
Meta's Muse agent arrives on Apple Watch and Wear OS
Meta's Muse AI agent is now available on Apple Watch and Wear OS, according to Beebom Gadgets. Shipping on both watch platforms at once signals that Meta treats wearables as a standard distribution surface for the agent rather than a single-platform experiment.
11/11, 21:57
Parloa and OpenAI expand partnership to push AI agents into customer experience
Parloa and OpenAI said on November 11 that they have expanded their partnership to advance AI agents for customer experience. The announcement did not disclose deal terms, duration or which models are involved, framing the move as a scale-up of existing work.
11/11, 19:58
KT says only 7 of Token Factory's 73 AI models run on its own hardware as it plans more GPUs
Tech Times reports that KT's Token Factory platform offers 73 AI models but runs only seven of them on its own hardware, while the company plans to buy more GPUs. The gap highlights the capital and time needed to build sovereign inference capacity, even for an operator that can aggregate models quickly through partnerships.