DeepMind CEO Demis Hassabis pulled the company's AGI timeline forward to 2029 in an Axios interview — a notably more aggressive public goal than other frontier-lab CEOs. He cited one or two remaining technical breakthroughs without naming them. The same interview revealed that DeepMind's multi-agent "Co-Scientist" system is now running across all 17 U.S. Department of Energy national laboratories

For the past six months, frontier-lab CEOs have said things like "sooner than you think" or "within a few years," but all of them avoided naming a specific year. Sam Altman's line has been "soon" without a date attached. Meanwhile DeepMind's Co-Scientist had been running in pilot settings at several research institutions, but the full DOE-portfolio scale was not publicly confirmed until now.

"AGI is coming" has been treated as an industry assumption, but a sitting frontier-lab CEO stating "2029" on the record is different — it turns a vague horizon into a planning constraint. If you're making hiring, budget, or technology-stack decisions that extend three to four years out, that number now belongs in the conversation. The DOE deployment signals that AI is becoming research infrastructure, not just a research tool. That said, nothing changes in your day-to-day workflow today.

Signed a strategic partnership with Fujitsu (announced 5/27). With NEC (first global partner, 4/30) and Hitachi (HMAX by Hitachi etc., 5/20) already in place, the three largest Japanese system integrators have now all aligned with Anthropic in the same quarter

Going into last month, NEC (late April) and Hitachi (mid-May) had signed with Anthropic, but Fujitsu — Japan's third mega-SI — was conspicuously absent. Having all three move within a single quarter signals that their enterprise AI strategies have shifted from observation to competitive positioning. For buyers, this means Claude deployment went from "we'd need to hire a developer" to "we can call our existing SI."

For enterprises in government, finance, and manufacturing — the traditional stomping grounds of Japanese SIs — Claude evaluation just moved from an API-first conversation to a vendor-relationship conversation. That makes it easier to write into a budget proposal. The catch: SI-led deployments favor stability over speed, so access to cutting-edge features typically lags direct API access by months. If you're a team that builds directly on the API, this shift is largely background noise for you.

Partnered with Cisco to put Codex at the center of an enterprise-engineering overhaul: scale AI-native software development inside Cisco, accelerate Cisco's "AI Defense" initiative, and automate defect remediation workflows

Cisco ships enormous volumes of networking software and dedicates significant engineering capacity to security patches and vulnerability remediation. Over the past few months Codex spread quickly among individual developers, but large-scale enterprise deployments with real security workflows attached had been sparse. Cisco bringing Codex into both its development pipeline and its AI Defense product line is a different tier of validation.

Cisco becoming a reference case means "Codex running at enterprise scale inside a security-sensitive environment" is now something you can point to in a vendor evaluation. If you're in procurement or security at a telco, network vendor, or technology firm, this makes internal proposals easier to defend. For individual engineers, it likely accelerates the pace at which companies roll out Codex access. If your firm is a Cisco partner, the AI Defense angle opens a new joint-offering conversation.

Cognition raised $1B+ at a $26.2B valuation ($25B pre-money), led by Lux Capital and General Catalyst with new participants Ribbit Capital, Atreides, and Layer Global. The valuation more than doubled in under nine months (from $10.2B), spotlighting investor appetite for AI coding agents

Devin launched in 2025 to significant hype, but early assessments pegged its autonomous success rate as limited compared to the "fully autonomous engineer" framing. Through 2026, Devin 2.0 added Interactive Planning and parallel sessions, and real-world success stories started accumulating. The previous round came in around summer 2025 at roughly $10B — doubling in under nine months is an unusually fast step-up.

A raise this size implies real enterprise revenue; investors at this stage want contracts, not prototypes. If your team has been deferring a decision on AI coding agents, this is a strong signal the category is past the early-adopter phase. That said, the headline valuation carries a large expectations premium, so run your own ROI test before buying in. Unlike Cursor or GitHub Copilot, Devin bills by autonomous work time (ACUs), which makes cost forecasting harder — and that has not changed.

Kiro Web launched in preview — a no-install, browser-based variant: pick a GitHub repo, describe what you want, and the agent runs from spec drafting through coding to opening a PR. It shares the same "steering" model as the existing Kiro IDE (VS Code-based) and Kiro CLI, so reviewers can focus on the PR surface

Kiro has existed as a VS Code-based desktop IDE, which meant anyone trying it needed to install and configure an extension first. It did not serve the use case of non-engineers requesting a quick change, or of developers wanting a CI/CD-style trigger without a local IDE open. Browser-based coding agents like GitHub Copilot Workspaces got there first, but Kiro Web differentiates with a spec-driven flow — it writes a requirements document before it writes code.

Removing the install step lowers the bar for non-engineers — PMs and QA can now open a browser, describe a bug fix, and receive a PR to review. The spec-first approach means the agent is less likely to start coding in the wrong direction. Still in preview, so reliability on complex repos needs to be validated before any critical-path use. Practical upside today is mostly limited to teams already in the AWS ecosystem.

Released Music v2: can switch genres mid-track while keeping composition and vocals coherent, and supports re-generating individual sections (intro/verse/chorus) without breaking the rest of the song. Trained entirely on licensed data to lower rights risk, and API pricing was cut by up to 50% as ElevenLabs squares off against Suno

ElevenLabs has been the default for voice — TTS and cloning — but Music v1 was outclassed by Suno in both quality and ease of use, cementing a split that held for over six months. Meanwhile Suno has been navigating copyright litigation tied to its training data, creating an opening for a competitor that could claim clean licensing. ElevenLabs had that claim; it just needed the model quality to catch up.

A 50% API price cut goes straight to the bottom line for any product that calls the music endpoint regularly. Section-level regeneration — swap the chorus, keep the verse — is the kind of fine-grained control commercial workflows need; it is closer to working with a producer than clicking a generate button. The licensed-data angle matters most to media companies and agencies where legal review would otherwise block deployment. For casual creators where Suno's UI feels more natural, the price drop alone may not be enough to switch.

Commercialization is accelerating: Kuaishou's Q1 2026 earnings showed Kling AI revenue up 300%+ YoY to ¥650M (≈$100M) for the quarter, pulling overall group revenue (¥3.37B) above consensus. Management put the AI unit's annualized run-rate at roughly $500M, signaling that generative video has entered a monetization phase

Through 2025, generative video — Sora, Kling, Runway and others — generated enormous press but thin answers to the question of whether people were actually paying. The common pattern was high trial usage that did not convert to recurring subscriptions, and quality that impressed in demos but required extra effort to embed in real production workflows. Proving sustained revenue at scale had been the missing piece across the whole category.

300% YoY growth and a $500M ARR run-rate is the most concrete evidence so far that generative video converts into recurring revenue, and it suggests the use cases driving it are repeat-purchase ones — social content and ad creative — not one-off experiments. That is a tailwind for Sora, Runway, and Veo as well; the market is real. The caveat that matters: Kling is operated by a Chinese company, and enterprise security policies may still block adoption regardless of quality or price.