Google Cloud Next 2026 · Ryan Edge 18 places - Where I'm eating in Vegas for Cloud Next 2026

A list of all of the restaurants that I want to try in Vegas during #GoogleCloudNext 2026

maps.app.goo.gl/GYBEvsypa3z...

Measuring token usage is the new measuring velocity.

Velocity taught teams to reward inflated estimates. Token usage will teach teams to reward verbosity.

Goodhart's Law doesn't care what era you're in.

opus 4.7 is a beauty
a fresh yet elegant take on something we've seen before
a new standard, a definite marker of a new era

(i haven't tried it yet)

Claude is broken for me this morning, but not so broken that it can’t advertise

Sounds like the Tesla/Musk bros 6 years ago.

All the money

Developer velocity was already a flawed, gameable metric—blind to quality and long-term costs. AI makes it worse: output spikes, effort looks smaller, and the real bottlenecks move to review, testing, and correctness.

I think Paradise is the best post-apocalyptic bunker show.

1. Paradise
2. Fallout
3. Silo

Better

I think Anthropic's pricing, especially for API usage, has gotten out of hand, and we are probably paying their legal fees.

Project Hail Mary is very good. That is all.

software factory more like slop shop

us: we are struggling to figure out the best way to use coding agents, we don't have clarity yet

everyone else: our team is moving at speeds unheard of, all our PRs are ai generated, we've cleared 6 years of backlog

man we must really suck huh

Go slow to go vast.

the discipline is the actual skill. i write specs before i write prompts, review every diff, run tests before merge. the model does the implementation. what i'm really doing is software architecture with a weird new team.

i really don't care about using AI to ship more stuff
it's really hard to come up with stuff worth shipping

i want to ship the same amount of stuff with higher quality both in product and code

If you are asking engineers to take on product thinking, planning, and risk assessment in addition to their technical work, name it. Define it. Compensate for it. Do not let it happen silently and then wonder why your team is burned out.

AI Made Writing Code Easier. It Made Being an Engineer Harder
www.ivanturkovic.com/2026/02/25/...
The baseline moved, and nobody sent a memo. AI made code cheaper to produce, but made reviewing, understanding, and maintaining it harder. The acceleration trap is real.

The Factory Model: How Coding Agents Changed Software Engineering Software engineering is not about writing code anymore. It is about building the factory that builds your software.

The Factory Model: How Coding Agents Changed Software Engineering
addyosmani.com/blog/factor...
You're no longer writing code—you're building the factory that builds the software. Specs become leverage, tests become mandatory, and verification is the bottleneck.

Automate repository tasks with GitHub Agentic Workflows   Build automations using coding agents in GitHub Actions to handle triage, documentation, code quality, and more.

Automate repository tasks with GitHub Agentic Workflows
github.blog/ai-and-ml/a...
GitHub brings coding agents into Actions—write workflow intent in Markdown, run it with guardrails. Triage, docs, testing, and code quality on autopilot. The next layer of CI/CD.

Code has always been the easy part
laughingmeme.org/2026/02/09/...
The cost of code is plunging toward zero. That's genuinely new. But code is the easy part? That's not new at all. The hard part has always been the system—the human-technology hybrid that delivers value and evolves.

Beyond agentic coding AI dev tooling can do better than chat interfaces

Beyond agentic coding
haskellforall.com/2026/02/bey...
Agentic coding promises productivity, but the data says otherwise. A compelling case for "calm" AI tools—semantic browsers, commit splitters, and focus modes—that work alongside you rather than demanding a conversation.

A Guide to Effective Prompt Engineering In this article, we will explore different prompting approaches, from simple zero-shot instructions to advanced chain-of-thought reasoning.

A Guide to Effective Prompt Engineering
blog.bytebytego.com/p/a-guide-t...
Solid primer on the core techniques—zero-shot, few-shot, chain-of-thought, role prompting, and prompt chaining. The key insight: clarity beats complexity.

Agentic Engineering Agentic Engineering is a disciplined approach to AI-assisted software development that emphasizes human oversight and engineering rigor, distinguishing it fr...

Agentic Engineering
addyosmani.com/blog/agenti...
Vibe coding ≠ agentic engineering. One is YOLO prototyping, the other is AI implementation with human oversight—with specs, tests, and code review. The terminology matters because the discipline matters.

The Eternal Promise: A History of Attempts to Eliminate Programmers
www.ivanturkovic.com/2026/01/22/...

COBOL, 4GLs, CASE tools, no-code—every decade promises to eliminate programmers. Every decade creates more of them. The hard part was never typing code - it was systems thinking

Good reads from February

Inside a staff engineer’s AI spec-driven development workflow · Luma There’s a new AI coding workflow every week. But it’s hard to tell what’s real versus just a flashy demo. The fastest way to get better isn’t reading another…

If you've been curious about how to write better specs, reduce context drift, or just get more consistent output from your AI tools, Augment is hosting a free virtual event on Friday on spec-driven development.

luma.com/b3u2al1c?tk...

Key insight: Expensive ≠ Quality. Kimi K2.5 is open-source and low-cost, and it was the top performer. The lesson isn't about price — it's about context quality and whether the model actually follows the specification.

Quality: Who did it well?

Kimi went beyond the specification on every dimension and was one of the few to get the edge cases right. Auggie shipped with a broken feature. Big Pickle looked complete on paper; 4 of 11 E2E tests would fail.

Completeness: Who actually finished?

6 of 9 entries finished the task. MiniMax got most of the way there. Gemini produced almost nothing. Big Pickle technically completed it, but 4 of 11 E2E tests crash at runtime. Finished ≠ Working.