Weekly AI Updates (What's New Wednesday)·April 2026·7:26:21

Weekly AI Updates — April 2026 digest (5 episodes)

Karan Rana Host, Weekly AI Updates · Akshat Host, Weekly AI Updates

The short version

OpenAI and Anthropic both pushed agent infrastructure this month: ChatGPT Workspace Agents, Claude Managed Agents, and Claude Routines all launched or expanded, aimed at running tasks autonomously across third-party tools
Anthropic teased an unreleased, extremely powerful model (Mythos, under Project Glasswing) that reportedly beats every public model but is being withheld over cybersecurity risk
Google pushed the open/mobile front with Gemma 4 (140 languages, phone-ready) and previewed a leaked enterprise chat interface positioned against Claude Cowork
Image generation crossed a real usability line: free ChatGPT Images 2 now handles dense text, multilingual posters, custom aspect ratios, and multi-panel comics reliably, closing the gap with Gemini/Nano Banana
Gamma's Claude MCP connector and Google's design.md files both push toward persistent, agent-readable brand/design systems instead of one-off prompts
AI-driven layoffs (estimated 73k-100k+ affected in 2026) and stunts like an AI agent running a real San Francisco retail store fueled recurring debate about AI's economic and workplace impact
Tools sighted this month: ChatGPT (Images 2, Workspace Agents, Codex, Chronicle), Claude (Managed Agents, Routines, Claude Code, Cowork, Claude Design, live artifacts), Gemini/Google (Vids, AI Studio, Stitch, Gemma 4, Colab Learn mode, AI Shopping), Gamma, Comic AI, Comics Maker AI, Pika Labs, Perplexity Computer, Microsoft Copilot, Cursor, Zapier, Lovable, Replit, Z.ai/GLM, Higgs Field, Netflix video-scene model, SpaceX Colossus

The concepts

Agentic AI moves from chat to autonomous execution

Across April, major labs pushed agents beyond single-turn chat into infrastructure that runs tasks autonomously and at scale. Anthropic's Claude Managed Agents let developers build once and deploy agents to many users without managing hosting; OpenAI's Workspace Agents, Claude Routines, and Codex Chronicle extend this into scheduled, cross-tool automation and persistent workflow memory.

Claude Managed Agents let developers build an agent once and have Anthropic host and scale it to many users via API, aimed largely at enterprise use (0:00:03, l3651436)

ChatGPT Workspace Agents can work across third-party tools like email and Slack, monitoring and executing workflows autonomously (0:17:16, l3651438)

Claude Routines, released the day before the session, let users schedule recurring AI tasks such as a 9am calendar and email briefing, replacing the older 'schedule' feature (0:34:20, l3651436)

OpenAI Codex added an opt-in 'Chronicle' feature that takes screenshots of a user's workflow to build persistent context memory; currently research preview only (0:29:28, l3651438)

An experimental AI agent named Luna was given a real 3-year retail lease in San Francisco and told to run the store profitably, including posting job listings and buying inventory (0:56:05, l3651437)

Cursor 3 and 'Cursor Anywhere' added support for running up to 8 parallel AI coding agents and executing workflows on remote machines (0:42:19, l3651435)

Frontier model race intensifies alongside safety caution

Multiple labs previewed or deliberately withheld extremely powerful models this month, citing both competitive pressure and safety risk. Anthropic's unreleased 'Mythos' model under Project Glasswing reportedly outperforms all public models and found a real exploitable vulnerability, while Google, SpaceX, and OpenAI raced out new specialized and general models.

Anthropic's unreleased Project Glasswing model, called Mythos, reportedly scores about 20-22% better than any public model and hit 78% on SWE-bench; only 50-100 vetted companies plus the US government have access (0:12:07, l3651436)

SpaceX is training a family of 7 specialized models (referred to as Colossus) instead of one general model, targeting domains like medicine and engineering (0:02:00, l3651436)

Google released Gemma 4 as a free, open, mobile-optimized model supporting 140 languages, positioned to run on-device for privacy-sensitive use cases (0:22:14, l3651436)

Independent researchers claim Anthropic quietly reduced Claude's reasoning effort (fewer reasoning passes) while keeping prices the same, to cut inference cost; Anthropic has not confirmed this and it remains a debate (0:10:06, l3651436)

DeepMind is bringing in philosophy and ethics experts to study what constrains harmful AI behavior, part of a broader industry safety push echoed by Anthropic's caution over Mythos (0:09:22, l3651437)

A research paper discussed on-air warns that AI-driven layoffs, estimated at 73,000 to over 100,000 jobs affected in 2026, could erode consumer demand faster than the economy can absorb displaced workers (0:41:44, l3651437)

AI image, comic, and design generation gets production-ready

Image-generation tools crossed a real usability threshold this month, reliably handling dense text, multi-panel comics, multilingual output, and brand-consistent design. Live demos moved from clunky third-party comic tools to Gemini and finally free ChatGPT Images 2, while Gamma's Claude MCP connector and Google's design.md files brought persistent style and brand control into agentic workflows.

Comic AI and Comics Maker AI could generate multi-panel comic strips with chosen art styles but frequently misspelled dialogue and misassigned speech to the wrong character (0:44:42, l3651434)

Gemini proved better at character-consistent multi-panel comics from an uploaded reference photo, including click-to-edit panel corrections via annotation (0:52:44, l3651434)

Gamma's new Claude MCP connector lets users generate full Gamma presentations, with chosen theme, tone, and layout, from inside a Claude chat without switching tabs (0:56:33, l3651435)

Google's design.md is a text file format storing a brand's colors, typography, and rules so agents (via Google Stitch) generate consistent, on-brand UI and design output instead of generic results (0:35:30, l3651438)

ChatGPT Images 2, on the free tier, reliably produces dense correctly-spelled text, multilingual output (demoed in Hindi), custom aspect ratios, and merged 4+ panel comic strips in a single image (0:47:37, l3651438)

ChatGPT Images 2 lacks a persistent character or asset library, so consistent characters only hold within a single generation thread, not across days or new chats (1:28:05, l3651438)

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.

01Agentic AI moves from chat to autonomous executionAcross April, major labs pushed agents beyond single-turn chat into infrastructure that runs tasks autonomo…›

Claude Managed Agents let developers build an agent once and have Anthropic host and scale it to many users via API, aimed largely at enterprise use (0:00:03, l3651436)

ChatGPT Workspace Agents can work across third-party tools like email and Slack, monitoring and executing workflows autonomously (0:17:16, l3651438)

Claude Routines, released the day before the session, let users schedule recurring AI tasks such as a 9am calendar and email briefing, replacing the older 'schedule' feature (0:34:20, l3651436)

OpenAI Codex added an opt-in 'Chronicle' feature that takes screenshots of a user's workflow to build persistent context memory; currently research preview only (0:29:28, l3651438)

An experimental AI agent named Luna was given a real 3-year retail lease in San Francisco and told to run the store profitably, including posting job listings and buying inventory (0:56:05, l3651437)

Cursor 3 and 'Cursor Anywhere' added support for running up to 8 parallel AI coding agents and executing workflows on remote machines (0:42:19, l3651435)

02Frontier model race intensifies alongside safety cautionMultiple labs previewed or deliberately withheld extremely powerful models this month, citing both competit…›

SpaceX is training a family of 7 specialized models (referred to as Colossus) instead of one general model, targeting domains like medicine and engineering (0:02:00, l3651436)

Google released Gemma 4 as a free, open, mobile-optimized model supporting 140 languages, positioned to run on-device for privacy-sensitive use cases (0:22:14, l3651436)

03AI image, comic, and design generation gets production-readyImage-generation tools crossed a real usability threshold this month, reliably handling dense text, multi-p…›

Comic AI and Comics Maker AI could generate multi-panel comic strips with chosen art styles but frequently misspelled dialogue and misassigned speech to the wrong character (0:44:42, l3651434)

Gemini proved better at character-consistent multi-panel comics from an uploaded reference photo, including click-to-edit panel corrections via annotation (0:52:44, l3651434)

Gamma's new Claude MCP connector lets users generate full Gamma presentations, with chosen theme, tone, and layout, from inside a Claude chat without switching tabs (0:56:33, l3651435)

ChatGPT Images 2 lacks a persistent character or asset library, so consistent characters only hold within a single generation thread, not across days or new chats (1:28:05, l3651438)

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
Claude Mitchell's / Claude Mitchell	Unclear/uncertain — transcript names a leaked, extremely powerful post-Opus Anthropic model tied to cybersecurity capability; exact codename as heard, not independently confirmed
TurboQuant	As-heard name for a Google Research algorithm claimed to cut AI hardware memory usage roughly 6x; exact product name uncertain
Zed AI / Zedai / Z AI (formerly Zipu AI)	Likely Z.ai (formerly Zhipu AI), the Chinese lab behind the GLM model family referenced as 'GLM 5v'
OpenClaw	Uncertain — used alongside 'Claude Code' and 'Cloud Code' as a separate open-source agent/orchestration tool; exact product name and vendor unclear from audio
Altair (software company, stock down from ~$165 to $71)	Likely mishearing of Atlassian, maker of Jira, cited as an example of AI disrupting SaaS software valuations
images 2.4 / Image 2	OpenAI's newly launched image generation model, referred to throughout as 'ChatGPT Images 2'; precise official name uncertain from audio

True on recording day — verify before relying

T
h
i
s
d
i
g
e
s
t
s
u
m
m
a
r
i
z
e
s
l
i
v
e
w
e
e
k
l
y
-
u
p
d
a
t
e
s
e
s
s
i
o
n
s
r
e
c
o
r
d
e
d
i
n
A
p
r
i
l
2
0
2
6
a
n
d
r
e
f
l
e
c
t
s
t
h
e
h
o
s
t
s
'
r
e
a
l
-
t
i
m
e
r
e
p
o
r
t
i
n
g
,
i
n
c
l
u
d
i
n
g
s
e
v
e
r
a
l
l
e
a
k
s
,
r
u
m
o
r
s
,
a
n
d
u
n
c
o
n
f
i
r
m
e
d
c
l
a
i
m
s
(
e
.
g
.
,
A
n
t
h
r
o
p
i
c
'
s
M
y
t
h
o
s
/
P
r
o
j
e
c
t
G
l
a
s
s
w
i
n
g
,
a
l
e
a
k
e
d
G
o
o
g
l
e
e
n
t
e
r
p
r
i
s
e
c
h
a
t
i
n
t
e
r
f
a
c
e
,
G
P
T
-
5
.
5
s
p
e
c
u
l
a
t
i
o
n
,
a
n
d
d
i
s
p
u
t
e
d
c
l
a
i
m
s
a
b
o
u
t
C
l
a
u
d
e
'
s
r
e
a
s
o
n
i
n
g
e
f
f
o
r
t
b
e
i
n
g
r
e
d
u
c
e
d
)
.
T
o
o
l
n
a
m
e
s
,
p
r
i
c
i
n
g
,
a
v
a
i
l
a
b
i
l
i
t
y
,
a
n
d
f
e
a
t
u
r
e
s
e
t
s
d
e
s
c
r
i
b
e
d
h
e
r
e
(
C
h
a
t
G
P
T
I
m
a
g
e
s
2
,
C
l
a
u
d
e
M
a
n
a
g
e
d
A
g
e
n
t
s
,
C
l
a
u
d
e
R
o
u
t
i
n
e
s
,
G
e
m
m
a
4
,
d
e
s
i
g
n
.
m
d
,
e
t
c
.
)
w
e
r
e
c
u
r
r
e
n
t
o
n
l
y
a
s
o
f
t
h
e
i
r
r
e
s
p
e
c
t
i
v
e
r
e
c
o
r
d
i
n
g
d
a
t
e
s
a
n
d
m
a
y
h
a
v
e
c
h
a
n
g
e
d
,
b
e
e
n
r
e
n
a
m
e
d
,
o
r
b
e
e
n
d
e
p
r
e
c
a
t
e
d
b
y
t
h
e
t
i
m
e
t
h
i
s
i
s
r
e
a
d
.
S
e
v
e
r
a
l
p
r
o
p
e
r
n
o
u
n
s
w
e
r
e
t
r
a
n
s
c
r
i
b
e
d
v
i
a
A
S
R
f
r
o
m
a
l
i
v
e
a
u
d
i
o
f
e
e
d
a
n
d
c
o
u
l
d
n
o
t
b
e
i
n
d
e
p
e
n
d
e
n
t
l
y
v
e
r
i
f
i
e
d
;
t
r
e
a
t
n
a
m
e
s
f
l
a
g
g
e
d
i
n
a
s
r
C
o
r
r
e
c
t
i
o
n
s
a
s
a
p
p
r
o
x
i
m
a
t
e
.
V
e
r
i
f
y
c
u
r
r
e
n
t
t
o
o
l
c
a
p
a
b
i
l
i
t
i
e
s
a
n
d
p
r
i
c
i
n
g
d
i
r
e
c
t
l
y
b
e
f
o
r
e
r
e
l
y
i
n
g
o
n
t
h
i
s
c
o
n
t
e
n
t
f
o
r
c
o
u
r
s
e
m
a
t
e
r
i
a
l
.