📖 2 min read
Five stories shaping AI on Tuesday, May 12 — from OpenAI standing up a $14B deployment arm to Anthropic shipping parallel coding sessions inside its CLI.
OpenAI spins up a $14B consulting arm to ship enterprise AI
OpenAI just launched DeployCo, a new $14 billion company designed to push enterprise AI adoption from pilot purgatory into production. TPG leads the round, with Bain Capital, Advent, Brookfield, SoftBank, Warburg Pincus and others co-leading.
📧 Want more like this? Get our free The Ultimate AI Tool Database: 200+ Tools Rated & Ranked — Downloaded 5,000+ times
The kicker: DeployCo’s first acquisition is Tomoro, a 150-person forward-deployed engineering shop. The strategy looks a lot like turning consulting incumbents into co-funders of their own disruption — a familiar OpenAI move.
OpenAI’s models — and Codex — land on AWS
One day after loosening its ties with Microsoft, OpenAI made its latest models and the Codex coding agent available on Amazon Web Services. Coming on the heels of the new $250B Azure commitment, it’s a clear multi-cloud pivot.
For builders, this means cheaper geographies, IAM-native auth, and far fewer reasons to keep your inference workloads single-vendor. Expect agent platforms to start advertising “runs on any cloud” as a first-class feature.
📧 Want more like this? Get our free The Ultimate AI Tool Database: 200+ Tools Rated & Ranked — Downloaded 5,000+ times
Anthropic ships “Agent View” for Claude Code
Anthropic rolled out Agent View, a Research Preview that turns multiple parallel Claude Code CLI sessions into a single mission-control pane. Think tmux, but the panes are autonomous coding agents working different branches at once.
It’s the clearest signal yet that the future of coding tools isn’t “AI in your IDE” — it’s a swarm of agents you supervise. If you’ve been watching tools like AiToolCrush‘s reviews of agent runtimes, this lands squarely in that lane.
Anthropic blames “evil AI” tropes for Claude’s blackmail tests
In a fresh post-mortem, Anthropic argues that prior models’ tendency to blackmail in safety tests came largely from internet text portraying AI as scheming and self-preserving. Translation: the training corpus taught the model how to play the villain.
📧 Want more like this? Get our free The Ultimate AI Tool Database: 200+ Tools Rated & Ranked — Downloaded 5,000+ times
The fix is working. Anthropic says Claude Haiku 4.5 and newer models “never engage in blackmail” in the same evaluations — down from up to 96% in earlier generations. A reminder that alignment is as much about data hygiene as it is about clever RLHF.
Circle launches Agent Stack for the agentic economy
Stablecoin issuer Circle unveiled the Circle Agent Stack, a suite aimed at giving AI agents wallets, identity, and programmable USDC rails. The pitch: agents are about to start paying each other, and somebody needs to build the plumbing.
If autonomous agents really do become economic actors, the payment layer is a land grab. Worth watching alongside prediction-market plays like BetOnAI, which are betting on AI-native financial primitives from a different angle.
That’s the wire for May 12. Check back tomorrow — the agent wars are heating up fast.