Weekly AI Updates (What's New Wednesday)·May 2026·6:32:14

Weekly AI Updates — May 2026 digest (4 episodes)

Karan Rana Host, Weekly AI Updates · Vishwas Girdar Guest Host, Weekly AI Updates (cinematographer/video editor, filled in for Karan Rana on l3836479 and l3836480)

▶local recording · LMS lesson

The short version

Claude Skills became the month's signature Anthropic feature: reusable, document-based workflows (skill.md + assets + references) invoked by slash command or trigger keyword, with a growing library of prebuilt templates.
Multi-agent orchestration went mainstream — Claude Console managed agents, Claude Code's new agent-view dashboard, and ByteDance's open-source Deerflow 2 all let users run teams of specialized sub-agents reporting to one master agent.
Video generation vendors raced toward agentic, physics-aware creative workflows: Grok Imagine's single-canvas agent mode, Google Omni's character/environment/style swaps, and Seedance 2/Kling as top-tier models.
Ambient, screen- and voice-aware assistants proliferated: Google's experimental AI pointer, Mira Murati's real-time multimodal AI, Meta AI's interruption-aware voice mode, and Google's Gemini Spark, Ask YouTube, and Ask Maps.
Compute and geopolitics shaped the AI supply chain — China blocked Meta's roughly $2B bid for Manus AI, and Anthropic leased about 220,000 NVIDIA chips from SpaceX's underused Colossus 1 cluster to fix Claude's speed and rate-limit complaints.
Everyday productivity tools absorbed AI natively: ChatGPT and Claude inside Excel/Google Sheets, Shopify-to-chat MCP integrations, and Perplexity's Computer agent landing inside Microsoft Teams.
Google IO (May 19-20) dominated the back half of the month, unveiling roughly 100 features across Search, YouTube, Maps, the Omni video model, Gemini Spark, and Samsung-partnered AI glasses.
Tools sighted this month: Claude (Skills, Console/Managed Agents, Code, Cowork, Dispatch), Grok Imagine, Google Gemini (3.5, Spark, Omni, magic pointer/cursor), NotebookLM, Runway, Perplexity Computer, Shopify MCP, ChatGPT (5.5, 5.5 Instant, Excel/Sheets add-in, Codex mobile), Kriya 2, Seedance 2 (ByteDance), Deerflow 2, Kling, Higgsfield, HeyGen, Neuralink, Manus AI, Google Flow, SynthID, Google Anti-Gravity, Numerous AI.

The concepts

Claude Skills and Multi-Agent Orchestration

Across the month, Anthropic's ecosystem shifted from single-shot prompting toward packaged, repeatable capability: Claude Skills bundle instructions, brand/reference assets, and resources into a reusable skill.md-based folder invoked by slash command or trigger keyword, while Claude Console/Code layered multiple specialized agents under one master agent with a shared dashboard.

Claude Skills package a repeatable workflow (an SOP plus assets and references) into a skill.md file invoked via slash command or trigger keywords in normal chat (0:49:39, l3836478)

The most popular prebuilt Claude skill is brand guidelines; skills can be created from a prebuilt template, a zipped folder upload, or by asking Claude to convert an existing chat/PDF into a skill (1:06:18, l3836478)

Skills can call each other (a design skill invoking a copywriting skill) the same way a master agent hands off to sub-agents, though nesting is intentionally shallow (0:56:09, l3836478)

Claude Console added managed agents/templates (e.g., a Deep Researcher preset) that run as specialized sub-agents under a master agent, billed from a separate prepaid token wallet rather than the normal subscription (0:39:27, l3836479)

Claude Code's new agent-view dashboard lets several specialized agents (sales, marketing, production) report to one master agent, all monitored and approved from a single screen (0:59:20, l3836480)

Anthropic leased roughly 220,000 NVIDIA chips from SpaceX's underused Colossus 1 supercomputer; the host credited the deal with fixing Claude's reported slowdowns and reduced rate limits (0:35:25, l3836479)

AI Video Generation Arms Race

Image/video model vendors converged on agentic, single-canvas creative workflows this month, adding storyboard planning, style/character/environment transfer, and physics-aware motion — moving beyond simple text-to-video toward director-like iterative control.

Grok Imagine's new agent mode turns a single chat canvas into a full creative workflow — brainstorm, generate/edit images, pick a model, then convert to video without leaving the chat (0:15:16, l3836478)

ByteDance's Seedance 2 (heard as "CDance"/"CDANS") is described as the most advanced and most expensive video model, paired in demos with GPT Image 2 for storyboard-to-video planning (0:25:22, l3836478)

Kriya 2 lets users prompt primarily with image references and saved mood boards instead of text, aimed at consistent style transfer for social and film assets (0:49:11, l3836480)

Google Omni (built on Gemini 3.5) launched as a combined video generation and editing model with character swap, environment swap, style swap, gravity/lighting-aware physics, and beat-synced audio-video generation (0:50:23, l3836481)

ByteDance also released Deerflow 2, a free open-source agent framework for building research/coding agents, positioned as a cheaper alternative to Claude-based automation (1:05:22, l3836480)

Ambient AI Assistants Everywhere

Multiple vendors previewed AI that lives ambiently on the screen, in voice, or across a product suite rather than in a dedicated chat window — reacting to cursor position, watching video in real time, or running proactively in the background across email, calendar, and search.

Google teased a screen-aware AI pointer/cursor (via Gemini, showcased at deepmind.google) that can summarize, translate, or act on whatever the cursor is touching on screen; not yet public (0:10:44, l3836480)

Mira Murati's Thinking Machines Lab previewed a real-time multimodal AI that talks, listens, and translates simultaneously at roughly 0.4-second latency (0:16:47, l3836480)

Meta AI added advanced voice conversations with interruption support and real-time vision, letting the assistant react naturally mid-sentence rather than restarting its answer (0:38:58, l3836480)

Google's Gemini Spark bundles Gmail, Calendar, Chat, and Tasks into one always-on personal assistant that keeps working in the cloud after the laptop is closed (0:28:04, l3836481)

Google IO also introduced Ask YouTube and Ask Maps, which answer natural-language queries by jumping straight to the relevant video timestamp or route/context (0:38:10, l3836481)

Every concept, three clicks deep

The same concepts as a quick reference: the closed row is the glance, open is the study card, and every timestamp jumps into the recording.

01Claude Skills and Multi-Agent OrchestrationAcross the month, Anthropic's ecosystem shifted from single-shot prompting toward packaged, repeatable capa…›

Claude Skills package a repeatable workflow (an SOP plus assets and references) into a skill.md file invoked via slash command or trigger keywords in normal chat (0:49:39, l3836478)

Skills can call each other (a design skill invoking a copywriting skill) the same way a master agent hands off to sub-agents, though nesting is intentionally shallow (0:56:09, l3836478)

Claude Code's new agent-view dashboard lets several specialized agents (sales, marketing, production) report to one master agent, all monitored and approved from a single screen (0:59:20, l3836480)

02AI Video Generation Arms RaceImage/video model vendors converged on agentic, single-canvas creative workflows this month, adding storybo…›

Kriya 2 lets users prompt primarily with image references and saved mood boards instead of text, aimed at consistent style transfer for social and film assets (0:49:11, l3836480)

ByteDance also released Deerflow 2, a free open-source agent framework for building research/coding agents, positioned as a cheaper alternative to Claude-based automation (1:05:22, l3836480)

03Ambient AI Assistants EverywhereMultiple vendors previewed AI that lives ambiently on the screen, in voice, or across a product suite rathe…›

Mira Murati's Thinking Machines Lab previewed a real-time multimodal AI that talks, listens, and translates simultaneously at roughly 0.4-second latency (0:16:47, l3836480)

Meta AI added advanced voice conversations with interruption support and real-time vision, letting the assistant react naturally mid-sentence rather than restarting its answer (0:38:58, l3836480)

Google's Gemini Spark bundles Gmail, Calendar, Chat, and Tasks into one always-on personal assistant that keeps working in the cloud after the laptop is closed (0:28:04, l3836481)

Google IO also introduced Ask YouTube and Ask Maps, which answer natural-language queries by jumping straight to the relevant video timestamp or route/context (0:38:10, l3836481)

Tools referenced

Tool	Coverage	Moment	Context

Action items

Extraction notes

This page was built from an auto-generated transcript, which garbles product and people's names. Those were corrected silently in everything above and logged here for transparency. The warnings flag claims that were true on the recording day but change fast.

Transcript corrections applied

The transcript says	The trainer actually means
CDance / CDANS / c dance	Seedance (ByteDance's video generation model, e.g. Seedance 2)
scale	Claude "skill" — host's audio consistently transcribes the Claude Skills feature as "scale"
cloud / cloud code / cloud co work	Claude / Claude Code / Claude Cowork (Anthropic's assistant products)
Antropek / anthropek dot com	Anthropic (anthropic.com)
Hicksfield / Higgsville / X field	Higgsfield (AI image/video generation platform)
Johnny Ives	Jony Ive, former Apple chief design officer, reportedly working with Sam Altman on an AI-first device

True on recording day — verify before relying

T
h
e
s
e
f
o
u
r
e
p
i
s
o
d
e
s
(
M
a
y
6
-
2
7
,
2
0
2
6
)
a
r
e
l
i
v
e
A
I
-
n
e
w
s
r
o
u
n
d
u
p
s
d
e
l
i
v
e
r
e
d
i
n
f
o
r
m
a
l
l
y
b
y
r
o
t
a
t
i
n
g
h
o
s
t
s
r
e
a
d
i
n
g
p
r
e
s
s
r
e
l
e
a
s
e
s
,
t
e
a
s
e
r
v
i
d
e
o
s
,
a
n
d
a
u
d
i
e
n
c
e
c
h
a
t
i
n
r
e
a
l
t
i
m
e
;
s
e
v
e
r
a
l
i
t
e
m
s
w
e
r
e
e
x
p
l
i
c
i
t
l
y
d
e
s
c
r
i
b
e
d
o
n
-
a
i
r
a
s
u
n
l
a
u
n
c
h
e
d
,
b
e
t
a
/
t
r
u
s
t
e
d
-
t
e
s
t
e
r
-
o
n
l
y
,
U
S
-
o
n
l
y
,
o
r
u
n
c
o
n
f
i
r
m
e
d
r
u
m
o
r
(
e
.
g
.
,
"
C
l
a
u
d
e
M
y
t
h
o
s
"
w
a
s
c
a
l
l
e
d
s
p
e
c
u
l
a
t
i
v
e
/
u
n
v
e
r
i
f
i
e
d
b
y
t
h
e
h
o
s
t
;
G
o
o
g
l
e
'
s
s
c
r
e
e
n
-
a
w
a
r
e
p
o
i
n
t
e
r
,
M
i
r
a
M
u
r
a
t
i
'
s
r
e
a
l
-
t
i
m
e
A
I
,
a
n
d
s
e
v
e
r
a
l
G
o
o
g
l
e
I
O
f
e
a
t
u
r
e
s
w
e
r
e
t
e
a
s
e
r
s
w
i
t
h
o
u
t
p
u
b
l
i
c
r
e
l
e
a
s
e
d
a
t
e
s
)
.
F
i
g
u
r
e
s
s
u
c
h
a
s
d
e
a
l
s
i
z
e
s
,
c
h
i
p
c
o
u
n
t
s
,
h
a
l
l
u
c
i
n
a
t
i
o
n
-
r
a
t
e
p
e
r
c
e
n
t
a
g
e
s
,
a
n
d
s
u
b
s
c
r
i
p
t
i
o
n
p
r
i
c
e
s
w
e
r
e
s
t
a
t
e
d
v
e
r
b
a
l
l
y
b
y
h
o
s
t
s
w
i
t
h
o
u
t
a
c
i
t
e
d
s
o
u
r
c
e
a
n
d
m
a
y
b
e
a
p
p
r
o
x
i
m
a
t
e
o
r
m
i
s
-
h
e
a
r
d
.
G
i
v
e
n
t
h
e
f
a
s
t
-
m
o
v
i
n
g
n
e
w
s
-
r
o
u
n
d
u
p
f
o
r
m
a
t
a
n
d
s
i
x
-
p
l
u
s
m
o
n
t
h
s
s
i
n
c
e
o
r
i
g
i
n
a
l
r
e
c
o
r
d
i
n
g
,
t
r
e
a
t
a
l
l
p
r
i
c
i
n
g
,
f
e
a
t
u
r
e
-
a
v
a
i
l
a
b
i
l
i
t
y
,
a
n
d
c
o
m
p
a
n
y
-
d
e
a
l
c
l
a
i
m
s
a
s
a
s
n
a
p
s
h
o
t
t
h
a
t
s
h
o
u
l
d
b
e
r
e
-
v
e
r
i
f
i
e
d
a
g
a
i
n
s
t
c
u
r
r
e
n
t
o
f
f
i
c
i
a
l
d
o
c
u
m
e
n
t
a
t
i
o
n
b
e
f
o
r
e
b
e
i
n
g
c
i
t
e
d
a
s
f
a
c
t
.