Shipping at Inference-Speed & The Future of Agentic Coding
Claude Code's creator shares his setup, Adobe's 2026 AI vision, and why human delight beats AI sludge
Happy New Year! I hope you had a restful break—or at least managed to sneak in some quality time with whatever projects you’ve been noodling on.
Today’s roundup is packed with AI coding workflows, creative tool updates, and some fascinating year-in-review pieces that’ll help you hit the ground running in 2026. Let’s dive in.
🤖 AI & Creative Tools
Boris Cherny’s Claude Code Setup
The creator of Claude Code shares how he actually uses his own tool. What strikes me most? He runs a surprisingly vanilla setup—proving that the best creative tools don’t need endless customization to be powerful.
Sometimes the simplest approach is the most effective. This is a masterclass in intentional tool use.
The Future of Agentic Coding: From Conductors to Orchestrators
Addy Osmani paints a compelling picture of where software development is headed. We’re moving from “writing code” to “conducting AI assistants” to eventually “orchestrating autonomous agent fleets.” It’s a shift that has massive implications for creative technologists.
The future isn’t about replacing developers—it’s about elevating what we can accomplish.
Shipping at Inference-Speed
This developer’s workflow has transformed so dramatically that they rarely read the code they produce anymore. Instead, they focus on high-level architecture while GPT-5.2 Codex handles implementation. It’s either inspiring or terrifying—maybe both.
We’re entering an era where shipping speed is limited by how fast you can think, not how fast you can type.
🎨 Creative Software & Design
How Adobe Thinks Creatives Will Use AI in 2026
Adobe’s VP of Product Marketing shares their vision: AI tools like Firefly won’t just help with single tasks—they’ll be woven throughout the entire creative workflow, from ideation through production. Whether you’re excited or skeptical, it’s worth understanding where the industry giants are betting.
The most interesting creative work in 2026 might be defining how we want AI to assist us.
Escaping AI Sludge: Why MVPs Should Be Delightful
AI-generated content is flooding the internet with generic, characterless products. This piece argues that delight and emotional design aren’t luxuries—they’re survival strategies in a world of AI sludge. Something to think about as we build.
In a sea of AI-generated sameness, human touch becomes your competitive advantage.
🔧 Tools & Resources
React Grab
A clever tool that bridges the gap between AI coding agents and your actual UI. Hover over any element, press a hotkey, and instantly give your AI assistant the context it needs. Small tool, big productivity boost.
Unscreen Pro
AI-powered video background removal that actually works. If you’re creating content, this one’s worth bookmarking.
SAFE-MCP: Security Framework for AI Agents
As AI agents get more capable, security becomes more critical. This Linux Foundation-backed framework provides standardized approaches to securing Model Context Protocol (MCP) ecosystems. Worth knowing about if you’re building with AI agents.
⚡ Quick Hits
2025: The Year in LLMs — Simon Willison’s comprehensive look back at LLM progress. Spoiler: coding agents and Chinese open-weight models were the big stories.
WRAP: Maximize GitHub Copilot — A framework for getting better results from AI coding assistants.
2026 Data Engineering Roadmap — ETL pipelines are out, “context systems” for AI agents are in.
Instagram CEO on Labeling Real Content — Mosseri proposes cryptographically signed images to prove authenticity.
21 Lessons From 14 Years at Google — Addy Osmani’s career wisdom. TL;DR: User obsession beats clever code.
That’s today’s roundup. I’m particularly curious what you think about the “shipping at inference-speed” concept—are we there yet, or is it still aspirational for most of us?
Hit reply and let me know what caught your eye. I read every response.
Until next time,
Macklin
The Creative Technologist is a daily newsletter exploring the intersection of creativity and technology. Subscribe here if this was forwarded to you.
Shipping at inference speed - this captures exactly what it feels like to build with autonomous agents now. I just shipped a complete Slay the Spire 2 card database in days, not weeks, because the iteration cycle compressed from hours to minutes.
The key insight: when agents can read code, understand context, make decisions, and execute changes autonomously, the constraint shifts from "how long does coding take" to "how quickly can I validate and decide." I could review and approve changes faster than traditional development cycles.
What's wild: the agent (Wiz) didn't just build what I asked for - it extended itself with new skills mid-project when it encountered gaps. Needed image processing? Created a skill. Needed deployment automation? Built it. This self-extension is what makes agentic coding fundamentally different from traditional tools.
I wrote about this rapid-build experience here: https://thoughts.jock.pl/p/slay-the-spire-2-everything-we-know-card-database-2026 - we're not just shipping faster, we're shipping at a fundamentally different operational tempo.
That Steipete piece about not reading code anymore resonated. I've been shipping with Claude Code at a similar pace and the bottleneck really is inference speed now, not typing. The Cerebras partnership with OpenAI is the most concrete move towards fixing that. Codex-Spark on wafer-scale silicon does 1,000+ tokens/s, which is fast enough that the tool stops feeling like a background worker and starts feeling like a pair programmer. Wrote a breakdown of the chip architecture and why it matters: https://reading.sh/chatgpt-and-codex-are-about-to-get-helluva-lot-faster-51ad25a7eed0