Dropbox bets $350M on agent storage, AWS ships coding-agent crews, and a 65B open model claims the U.S. coding lead

Nov 5, 2026

DEALS: Dropbox pays $350M for Tigris Data to build DBX Agent Cloud; Sentra raises $150M and Sand.ai $200M

Dropbox announced on Nov 5 that it is acquiring Tigris Data for $350 million, its largest acquisition to date. The deal is expected to close in Q4 2026, and Dropbox says it does not plan headcount cuts. Tigris, built by ex-Uber storage engineers who raised a $25 million Series A last year, will power DBX Agent Cloud: S3-compatible object storage aimed at teams building AI models and agents, with filesystem features coming later so agents can manage files across sessions. For AI coders, this is the clearest signal yet that agent-grade storage is a product category, not a side feature. Also in the same window: Sentra raised $150 million in a Series C led by Capital One Ventures (total funding now above $260 million) for agentic data security (Sentra's own post), and Sand.ai raised $200 million in a Series A+ on Nov 5 co-led by Cornerstone Capital and Wangsu, bringing its 2026 haul to about $300 million. Chinese media have reported a valuation above $1 billion for Sand.ai; the exact figure was not disclosed, so treat that as unconfirmed.

Source: https://techcrunch.com/2026/11/05/dropbox-acquires-tigris-data-for-350m-to-bolster-its-ai-iinfrastructure/

Nov 5, 2026

AGENTS: AWS ships Kiro bots so you manage coding-agent crews, not individual sessions

AWS made Kiro bots generally available in Kiro Crew on Nov 5. Kiro Crew is its open-source multi-agent environment; bots are named, long-running agents with their own memory (Notes), tools, permissions and ongoing chat. You launch parallel worker sessions from a bot, monitor them, redirect them and consolidate the results, and you can schedule recurring work such as overnight CI or Friday release notes, with dashboards to watch the fleet. The practical shift for AI coders is from babysitting one session at a time to managing a small crew. InfoWorld notes analysts warning about a "coordination tax" on tokens as you add more bots, so start with a few high-value recurring jobs rather than spinning up a swarm on day one.

Source: https://kiro.dev/blog/crew-bots/

Nov 4, 2026

OPEN WEIGHTS: Ricursive's Ripple-65B Preview claims U.S. open-weight lead on agentic coding

Ricursive released Ripple-65B Preview on Nov 4 (Apache 2.0), a dense 65-billion-parameter model trained on 12 trillion tokens with up to 1 million tokens of context. Weights are on Hugging Face, and it is also available through Together, GMI Cloud, OpenRouter and Fireworks. On the company's self-reported Terminal-Bench 2.1 numbers it scores 80.52 / 80.5, ahead of Reflection Beam 501B (80.1), Meta Muse Spark Open 816B (76.4), Inkling 975B (63.8) and Nemotron 3 Ultra 550B (56.4). Ricursive says the full R&D cost was $32 million. Treat the benchmarks as company-reported; the model still trails larger mixture-of-experts systems on some other suites such as DeepSWE. VentureBeat frames it as the current U.S. open-weight claim on agentic coding. For AI coders who want an Apache-licensed dense model they can host or rent, this is the one to try this week.

Source: https://www.ricursive.com/ripple

Nov 5, 2026

MODELS: Ant Ling's Ling-3.2-flash sees images and video, with a two-week free window

Ant Ling shipped Ling-3.2-flash on Nov 5, a 560-billion-parameter mixture-of-experts model with 25 billion active parameters, a 1-million-token context window and native image, text and video input (model page). Ant Ling positions it for complex visual understanding, multimodal long-horizon agents, 3D understanding and interaction, and medical multimodal work. It exposes OpenAI-compatible and Anthropic-compatible protocol APIs, and the first two weeks are free. No weights have been announced. Demos of Blender and 3D workflows have circulated on X; stick to the changelog for hard claims until Ant Ling publishes them on its own docs. Pricing after the free window is not yet on the public pricing table (third-party routers such as OrcaRouter have noted this). For indie game devs who need vision plus long context without committing spend yet, the free window is the try-it slot.

Source: https://developer.ant-ling.com/en/docs/getting-started/changelog

Nov 4, 2026

AGENTS: Oumi's FAiE now builds agents and retrains the models behind them

Oumi launched an expanded Frontier AI Engineer (FAiE) on Nov 4. FAiE now covers end-to-end agent development: tools, environments, eval, training, deploy and improvement from production failures (Oumi's product post). You describe a task in plain language and FAiE automates the engineering work behind approval gates. Bring Your Own Harness lets you keep your existing agent runtime. Training adds DPO, with GRPO in private preview. SiliconANGLE notes the loop from "build the agent" to "retrain the model that powers it" as the new pitch. For AI coders who already ship agents and then hit eval cliffs, the interesting part is the continuous-improve path from production failures, not another greenfield builder.

Source: https://www.businesswire.com/news/home/20261104154139/en/Oumi-Launches-the-Agent-That-Builds-Trains-and-Continuously-Improves-AI-Agents

Nov 2026

RESEARCH: Exo lets coding agents safely edit their own harness mid-task

A new arXiv paper from Krentsel, Ratnasamy, Stoica, Casado (a16z) and Goyal (Braintrust) introduces Exo, an open-source harness that lets coding agents edit their own runtime mid-task under a protected mechanism plus an editable policy, with a guardian that activates revisions (GitHub). On AppWorld, SREGym and open-world game tasks the authors claim 12–61% lower model cost versus controls while matching completion rates. For AI coders building agent runtimes, the useful idea is separating a locked safety/mechanism layer from a policy layer the agent is allowed to rewrite, instead of freezing the whole harness or letting the agent rewrite everything. The code is open source, so you can inspect the guardian path before you trust it on production work.

Source: https://arxiv.org/abs/2611.00463

Ship faster this week

Grab the free Agentdex starter kit with prompts and agent configs for Cursor, Claude Code and Codex.