AI News — July 12, 2026

Coding agents

• [2026-07-10] Cursor — Cursor 3.11 adds “side chats”: durable, parallel agent conversations that branch off the main thread. Spin one up with /side, /btw, or the plus button to ask questions, explore ideas, or investigate a tangent without interrupting the main agent conversation; each side chat inherits context from the main thread, is a full agent conversation in its own right, and can be revisited later or at-mentioned to pull its findings back into the main thread. The same release makes agent transcripts searchable and simplifies the project and repo pickers. It matters because it targets a specific friction in long agent sessions — the cost of derailing a running task to chase a side question — turning those detours into first-class, resumable conversations rather than lost context. (official)

  In practice this formalizes a workaround people already improvise — opening a second agent window (which starts cold) or interrupting the running one to ask “wait, what does this helper do” — into a branch that starts warm from the main thread and whose answer stays addressable by @-mention afterward.

  For Software Developers: You’re mid-refactor and spot a suspicious auth helper; instead of stopping the refactor agent to chase it, you open a /side chat that already carries the codebase context, dig into the helper there, then @-mention that side chat back into the main thread so the refactor resumes with the finding — no re-explaining where you left off.

• [2026-07-10] Kiro — MCP OAuth extended to strict enterprise servers, plus hooks on agent-driven file changes. Kiro’s latest changelog expands MCP OAuth to servers with stricter requirements (e.g. Figma): you can now set a client secret, point the redirect URI at a custom callback path, and skip Dynamic Client Registration by bringing your own pre-registered client ID, with new /mcp auth, /mcp cancel-auth, and /mcp logout commands to manage stored credentials. The release also brings hooks to agent-driven file changes (a v1 JSON format under .kiro/hooks/ with triggers for file events, tool-use interception, spec tasks, and prompt submission) and hot-reloads agent and MCP config on save so edits to mcp.json take effect without restarting the session — only affected servers restart. It matters because pre-registered OAuth apps and client secrets are exactly what enterprise identity teams require before wiring an agent into internal MCP servers. (official)

  Skipping Dynamic Client Registration is the load-bearing part: most enterprise IdPs disable DCR outright, so before this, pointing Kiro at an internal MCP server behind Okta or Entra simply couldn’t authenticate — a bring-your-own client ID plus a custom callback path is the flow those servers actually permit.

  For Solution Architects: You can now fold Kiro into the same app-registration model as the rest of your stack — register it as a confidential client in your IdP, point the redirect at your custom callback path, hand it the client secret, and skip DCR — so an internal MCP server (a Figma bridge, a private tools server) authenticates it the way your other first-party apps do, with no DCR exception carved out just for the agent.

• [2026-07-11] Anthropic — Claude Code enables auto mode by default on the enterprise cloud backends. Auto mode now runs without the CLAUDE_CODE_ENABLE_AUTO_MODE opt-in on Bedrock, Vertex AI, and Foundry, Bedrock is updated to Claude Opus 4.8, and the release fixes terminal freezes on long streamed output plus a batch of subagent, worktree, and credential bugs. A routine changelog bump, but the auto-mode default lands the same model-routing behavior enterprise users already had on the first-party API onto the three managed clouds. (official)

  For teams pinned to a managed cloud for compliance reasons, the quieter line is Opus 4.8 arriving on Bedrock — auto mode defaulting on just means dropping the CLAUDE_CODE_ENABLE_AUTO_MODE flag from your deployment configs, but the model bump is what closes the capability gap those teams had been running behind the first-party API.

Watch list

• Fable 5 steady-state cost cutover (tonight) — Anthropic’s Fable 5 stays included at up to 50% of weekly limits on Pro, Max, Team, and eligible Enterprise seats only through 11:59pm PT July 12; after tonight, access moves to metered usage credits at $10/$50 per million input/output tokens. This is the cutover the last several briefings flagged — from tomorrow, any path still routed to Fable 5 meters against a per-token budget rather than the included window. Watch whether standard Enterprise seats can enable usage credits at all. (official, prior coverage)

  Beyond tomorrow’s bill, this is the first included-then-metered transition Anthropic has run on a flagship, so how it lands doubles as the template for how the next model gets sunset off the included tier.

• Gemini 3.5 Pro (GA) — still a limited Vertex AI preview past the midpoint of July, with no model card, published benchmarks, or pricing for the promised 2M-token context and Deep Think mode. Reporting continues to point to a July 17 target tied to a full architectural rebuild, with early-tester complaints about token efficiency and coding cited for the slip; Google has confirmed none of it. With Grok 4.5 and Muse Spark 1.1 both shipped, it is the only major summer flagship still entirely unlaunched. Watch for whether even a model card lands before the 17th. (unconfirmed) (prior coverage)

  What makes the slip worth tracking is scope: a 2M-token, Deep-Think GA wouldn’t just add another flagship, it would reset the context-window and reasoning baseline the rest of the field is currently priced against.

• GPT-5.6 first independent benchmarks — with GA rollout now broad, Sol/Terra/Luna are in wide use but the only public performance numbers remain OpenAI’s own. Muse Spark 1.1, Grok 4.5, and Opus 4.8 now carry published scores on the same coding suites (SWE-Bench Pro, Terminal-Bench, DeepSWE); watch for the first third-party eval that runs those against Sol. (prior coverage)

  Worth watching because every performance figure on Sol so far traces back to OpenAI, and independent SWE-Bench runs have landed below vendor claims often enough that the first neutral number, not the launch deck, is what will actually size it.

• MCP spec finalization (July 28) — the 2026-07-28 release candidate (stateless core, Extensions framework, Tasks, MCP Apps, authorization hardening) is in its validation window with the Python/TypeScript/Go/C# SDKs updating against it in beta. Watch for the remaining SDK support landing and any last changes to the new Mcp-Method/Mcp-Name HTTP headers that security analysts have flagged as fresh attack surface. (official, prior coverage)

  What raises the stakes on those header changes is that the stateless-core rework is breaking for existing servers, so any last-minute Mcp-Method/Mcp-Name adjustment lands as migration work for every SDK consumer, not just a spec footnote.