Key Highlights

  • The I/O 2026 dust settles into a single story: agents, not models. The Rundown’s day-after recap frames Google’s whole keynote as a “deploy Gemini as an agentic engine” play across Search, Workspace, and the new Spark personal agent — with Omni at 4× speed and roughly half the cost of competing video generators.
  • NVIDIA quietly shipped the most interesting agent-safety primitive of the week. Its new Verified Agent Skills program treats agent capabilities like signed software packages — daily scans by a new “SkillSpector” tool, cryptographic signing, and skill cards documenting provenance — pushing trust down from runtime guardrails to the capability layer itself.
  • Slack is positioning DMs as the agent-to-agent protocol. Stack Overflow’s podcast with Slack CPO Jaime DeLanghe argues the enterprise chat substrate already solves the hardest agent-interop problems (identity, context, audit) and that bots, not new APIs, will be how agents talk to each other.
  • Pushback on “AI-generated” accusations is starting. A widely-shared Lobsters post — “LLemdashes” — flips the usual concern: dismissing real writers by emdash-spotting silences entry-level authors at exactly the moment AI tools are already suppressing their wages. Worth reading as a counterweight to the current detector arms race.

Analysis & Opinion

Gemini’s busy agentic day at Google I/O — The Rundown

The Rundown’s day-after framing of I/O 2026 cuts through the firehose: Google’s central bet isn’t on a flagship model but on making Gemini “capable, fast, and affordable enough” to be the default agentic engine inside products people already use. Gemini Omni gets singled out for the price/performance combination — text/image/audio/video inputs to video output at 4× competitor speed and roughly half the cost. The redesigned Search (cross-modal input, 24/7 information agents, generative UI) is treated as the most consequential reveal because it’s the surface that touches the most users daily.

LLemdashes — Lobsters

A pointed defense of human writers being misidentified as AI on the basis of stylistic markers like the emdash. The author concedes the legitimate concern — “generated text, assembled by nobody and increasingly for nobody, should be dismissed” — but argues the social cost of false positives is being borne by exactly the writers least able to defend themselves: entry-level workers, whose wages and career advancement are already being suppressed by AI tooling. The piece is short on solutions but useful as a mirror for anyone who has reflexively typed “this reads AI-generated” into a comment thread this year.

New Products & Tools

Accelerating AI impact in Singapore — Google

Google announced an expanded national AI partnership with Singapore’s Ministry of Digital Development and Information, organized around three pillars: addressing societal challenges, workforce readiness, and enterprise AI enablement with explicit trust-and-safety scope. Google DeepMind’s Singapore lab will anchor the research side under the company’s “National Partnerships for AI” framework.

Running Guide agent: a step towards running unbounded — Google DeepMind

A chest-mounted Pixel 10 Pro plus a Gemma 4-powered scene-understanding pipeline lets blind and low-vision runners train independently — without a sighted guide or fixed track. The architecture is interesting on its own terms: local on-device inference handles urgent safety alerts (collision, drop-off), while richer scene reasoning runs in the cloud. A planner module negotiates the run plan and pulls weather and map data, and a coaching module separates audio cues into danger/warning/notice tiers. SG Enable is running field trials and smart-glasses form factors are next.

Pack your agentic stack in Slack — Stack Overflow Blog

Stack Overflow’s podcast with Slack CPO Jaime DeLanghe argues Slack is the natural substrate for enterprise agents because the hard problems — identity, context inheritance, audit, human-in-the-loop — are already solved by the chat product. The provocative claim from the episode: “the best agent-to-agent protocol might be a DM.” Slack Dev Day announcements aim to make agents first-class citizens alongside the existing bot framework.

NVIDIA-Verified Agent Skills — NVIDIA Developer

NVIDIA introduced verified agent skills — portable capability bundles with provenance metadata, security scanning, and cryptographic signatures. A new internal tool, SkillSpector, evaluates both conventional software risks and agent-specific vulnerabilities before a skill enters the public catalog. Skills are catalogued daily from product teams, signed, and shipped with “skill cards” documenting origin and modifications. This is the most concrete attempt yet to move agent trust from runtime guardrails to the capability layer — closer to how Linux distros sign packages than how today’s agent frameworks vet tools.

Research

Mastering Agentic Techniques: AI Agent Evaluation — NVIDIA Developer

A practical framing of why model benchmarks (MMLU and friends) miss what matters for production agents. The recommendation set is short and load-bearing: prioritize task success over accuracy, log complete trajectories, treat tool-call correctness as a first-class metric, separately score reasoning soundness from computational efficiency, and instrument observability from day one.

NVIDIA and Google Cloud Empower the Next Wave of AI Builders — NVIDIA

The joint NVIDIA/Google Cloud developer community launched at last year’s I/O has crossed 100,000 members. New learning paths cover JAX on NVIDIA GPUs and a codelab on NVIDIA Dynamo for inference optimization — relevant if you’re moving RAG or agent workloads onto GKE with GPU acceleration.

References

  1. The Rundown, “Gemini’s busy agentic day at Google I/O,” The Rundown AI, 2026-05-20 [blog]
  2. wil.to, “LLemdashes,” via Lobsters, 2026-05-20 [blog]
  3. Google, “Accelerating AI impact in Singapore,” The Keyword, 2026-05-20 [blog]
  4. Google DeepMind, “Running Guide agent: A step towards running unbounded,” Innovation & AI, 2026-05-20 [blog]
  5. Ryan Donovan (with Jaime DeLanghe), “Pack your agentic stack in Slack,” Stack Overflow Blog, 2026-05-20 [blog]
  6. NVIDIA, “NVIDIA-Verified Agent Skills Provide Capability Governance for AI Agents,” NVIDIA Developer Blog, 2026-05-19 [blog]
  7. NVIDIA, “Mastering Agentic Techniques: AI Agent Evaluation,” NVIDIA Developer Blog, 2026-05-19 [blog]
  8. NVIDIA, “NVIDIA and Google Cloud Empower the Next Wave of AI Builders,” NVIDIA Blog, 2026-05-19 [blog]