Skip to content
All issues
Published issue

🍻 The Creative Stack Grew Hands

Finally, the art department gets the first round. Adobe turned Firefly toward full production with one workspace for image, video, assets, timelines, music, speech, and batch jobs. Figma gave responsive typography two small but delicious controls, then made approved skills and templates easier for teams to find. The week's bigger shift: creative AI is leaving the prompt box and learning how real work moves from taste to system to handoff.

01

Quick Pour

Friday in five minutes

🍻 The prompt box is becoming a production table.

Adobe’s August Firefly update puts image and video generation, project assets, a video timeline, music, speech, and batch interfaces in one production-shaped workspace. Figma added responsive text wrapping and organization-recommended creative resources. OpenAI’s mobile agent controls still matter, but they are the side pour this week. The main lesson is visual: creative AI gets useful when taste, repeatability, direct editing, and handoff survive past the first pretty output.

Covered first-party changes · August 14–20
MonTueWedThuFriSatSun
Aug 100000110
Aug 170210000
Aug 14
Figma shipped Balance and Pretty responsive text wrapping
Aug 17
Figma added organization-recommended skills, templates, libraries, and Make kits
Aug 18
OpenAI released ChatGPT for iOS 1.2026.223
Aug 18
OpenAI released Codex CLI 0.148.0 with export, forking, hooks, and reliability fixes
Aug 19
Adobe updated Firefly with a unified creative workspace, batch interfaces, music, and speech

Only covered first-party product changes from August 14–20 appear in this grid—not every AI release. Figma’s August 13 launch of 50+ Community agent skills is used only as direct context for its in-window August 17 resource-recommendation update. The DALL·E GPT deadline and skill spotlight are current official context outside the release grid.

02

This Week’s Flight

Lead pour · The prompt box grew a production floor

Firefly stopped auditioning and joined the crew.

Adobe’s August update puts image and video generation, project assets, and a video timeline in one Firefly workspace. It also adds interfaces for repeatable batch jobs with up to 500 input assets, plus generated music and expressive speech. That does not make the work finished or automatically on-brand. It changes the handoff: teams can now encode a repeatable production path, then spend human judgment on the frames, cuts, sound, rights, and exceptions that deserve it.

See Adobe Firefly’s August 2026 updates (opens in a new tab)

Design detail · Tiny control, immediate polish

Figma killed the lonely word.

Figma now supports two responsive text-wrap options across text layers, styles, and individual paragraphs. Balance distributes a headline more evenly across lines. Pretty keeps one orphaned word from dangling on the final line. This is not a flashy model demo; it is better. One deliberate typographic choice can now survive Figma Design, Slides, Buzz, and Sites without a designer hand-fixing every frame.

See Figma’s August 14 text-wrap release (opens in a new tab)

Taste system · Put the good tools on the top shelf

Design taste got a checkmark.

Figma admins can now recommend specific skills, templates, libraries, and Make kits to a workspace or organization. Those resources receive a visible recommendation mark, so the team can find the approved starting point instead of guessing. The timing matters: one day before this issue’s source window, Figma also opened a Community library with 50+ agent skills and tools to create and publish your own. The practical move is not ‘let AI design.’ It is curating a small, reviewable shelf of how your team designs.

See Figma’s August 17 resource update and August 13 skills context (opens in a new tab)

Side pour · The agent still left the desk

Your phone became an approval surface.

ChatGPT for iOS 1.2026.223 added direct Codex Remote launch, standard MCP forms, and editable message approvals. OpenAI also says canceled or edited steering messages no longer slip through while pairing, reconnection, task loading, and diff review received reliability work. Useful, yes—but this week it belongs beside the creative story, not on top of it. Treat the phone as a decision surface, never as permission to approve blind.

Read OpenAI’s ChatGPT and Codex changelog (opens in a new tab)
03

By the Numbers

Production scale · Capacity is not quality

One interface can now start with 500 assets.

Adobe says Firefly Creative Production interfaces can run batch jobs with up to 500 input assets. Against a one-file manual baseline, that is a 500-times larger starting queue. It is not a claim of 500-times faster work, perfect output, or unattended brand safety. It is a loud signal to make the recipe, exception path, and quality check explicit before scaling it.

Creative production job input capacity · Assets per run · Adobe maximum versus a one-file editorial baseline
One-file manual baseline
1 asset
Firefly interface ceiling
Up to 500 assets

500× the input count of a one-file baseline—not 500× faster, cheaper, or better.

Adobe documents a maximum of 500 input assets for Firefly Creative Production interfaces. Studio Tokyo supplies the one-file baseline only to visualize scale. Actual eligibility, duration, credits, outputs, review effort, and quality depend on the workflow and assets.

04

Smokin’ Hot Skill

Hot Skill · Ideal useAnthropic-maintained · Hosted by GitHub · Not paid

Anthropic Skills catalog / Skill on tap

Frontend Design

Use this when an AI-built interface keeps arriving in the same tasteful-but-generic costume. Anthropic’s official skill asks for one concrete subject, a deliberate visual direction, a compact token system, a memorable signature element, and a critique before code. Its sharpest move is naming the current AI-design defaults so the team can choose something specific instead of accidentally repeating them.

Ideal use
Push one branded landing page or product surface beyond generic AI styling
Give it
A real subject, audience, single page job, brand voice, assets, and visual references
Ask for
A named direction, color and type tokens, one signature move, and a self-critique

The skill is current official context outside the August 14–20 release grid. It is a critique lens, not a substitute for design judgment or the Studio Tokyo system. Anthropic-maintained, hosted on GitHub, and selected independently by Studio Tokyo. Not paid. This issue does not install or execute it.

Open the skill (opens in a new tab)
05

Try This Monday

One controlled move · 60 minutes

Make one idea survive three glasses.

Take one visual direction from hero to social crop to email banner. The goal is not maximum generation. It is finding which parts of the taste can become a system without sanding off the good weirdness.

  • Start with one visual plate, one seven-word headline, and four locked brand decisions: palette, type pair, logo treatment, and the thing the AI must not smooth out.
  • Build desktop hero, portrait social, and email-banner frames. In Figma, test Balance against Pretty on the same headline. Keep the option that preserves rhythm at all three widths instead of hand-tuning each line break.
  • Write a tiny production recipe for the repeated work: input asset, crop rule, background behavior, output size, and human check. Run three assets—not 500. Keep the system only if it reduces correction time without making the campaign look obedient and dead.
06

Last Call

Creative last call · August 30

The official DALL·E GPT closes August 30.

OpenAI says it will retire the official DALL·E GPT in ChatGPT on August 30. Download any images you want to keep before the cutoff and move active creation or editing workflows to ChatGPT Images. OpenAI says user-created GPTs with image generation are not affected. The announcement was published July 31; it is here because the operational deadline is now.

Read OpenAI’s DALL·E GPT retirement notice (opens in a new tab)