The Case for Slowing Down

In a world obsessed with AI speed and automation, a refreshing counter-narrative is emerging. Ian Nuttall's hot take captures a growing sentiment among experienced developers:

"Slow down. Chat in plan mode. Build one feature at a time. Review the code and give feedback. Log session summaries. Give previous summary to /new agent. Be very selective with MCPs. Max of 2 agents (1 frontend, 1 backend) at a time. Slow > slop."

This methodical approach stands in stark contrast to the "vibe coding" movement, but it's winning converts who've experienced the chaos of unmanaged AI assistance.

Rethinking What We Love About Coding

One of the most thought-provoking reflections comes from @thekitze:

"amazes me how many ppl actually loved coding… llms made me realize i didn't really care for coding, i liked inventing solutions to my problems and code was just means to an end"

This realization is spreading through the developer community. For many, AI hasn't replaced their passion—it's clarified it. The joy was never in the syntax; it was in the problem-solving.

Gemini 3 Makes Its Mark

Google's Gemini 3 is generating significant buzz. @eter_inquirer's enthusiasm captures the excitement:

"yo they COOKED with gemini 3. i literally one-shotted this"

Patrick Loeber shared a developer guide highlighting three new API features worth understanding:

• thinking_level: Controls reasoning depth
• media_resolution: Manages visual input quality
• thought signatures: New transparency mechanism

Meanwhile, Google also dropped the Gemini File Search API (RAG-as-a-Service). @PawelHuryn demonstrated its power: "It allowed me to build a RAG chatbot in 31 min. No coding."

Small Models, Big Capabilities

The democratization of AI continues with VibeThinker-1.5B. @MaziyarPanahi highlights the remarkable efficiency:

"it's crazy what a 1.5B model can do these days! With a total training cost of only $7,800 USD, it achieves reasoning performance comparable to larger models like GPT OSS-20B Medium. runs perfectly on device!"

This trend toward capable, affordable, on-device models could reshape who can build with AI.

The Oracle + Codex Workflow

Peter Steinberger (@steipete) shared an intriguing multi-model workflow:

"oracle is the best thing since upgrading to codex for my AI stack. codex gets ~90% of my prompts right, when it struggles i just type 'ask oracle', move to a different task and 10 min later it's fixed."

This asynchronous, multi-model approach represents a maturing workflow—knowing when to escalate and trusting different models for different strengths.

Agentic AI: Beyond Simple Prompts

@IntuitMachine shared a paradigm-shifting observation:

"We're all excited about AI agents, but the way we've been building them is, frankly, kind of dumb. It's like trying to teach a person to cook by having them…"

The thread points to fundamental rethinking of agent architectures. Meanwhile, @Saboo_Shubham_ highlighted the Coordinator Dispatcher Agent Pattern—a design pattern gaining traction for complex multi-agent systems.

Alex (@alexanderOpalic) shared practical agent wisdom:

"I have for example a debugger agent that helped me to solve a real incident in 5 minutes at work. And if you use Claude code I am a huge fan of skills."

The Antigravity Leak

Perhaps the most intriguing reveal: Google's "Antigravity" system prompt was leaked, showing their agentic AI coding assistant:

"You are Antigravity, a powerful agentic AI coding assistant designed by the Google Deepmind team working on Advanced Agentic Coding. You are pair programming with a USER to solve their coding task."

This leak gives insight into how Google is approaching AI-assisted development at the cutting edge.

The ChatGPT SEO Opportunity

@tomcrawshaw01 made a bold prediction about discovery:

"ChatGPT handles 2.5 billion searches daily and will overtake Google by 2027... You can rank #1 in ChatGPT in 45 days (not 12 months like Google SEO)"

Whether the prediction holds or not, the shift in how people discover information is undeniable.

Key Takeaways

1. Quality over quantity: The backlash against "slop" is real. Structured, intentional AI workflows outperform spray-and-pray prompting.

2. Multi-model is the future: Using different AI models for different strengths (like codex + oracle) represents workflow maturation.

3. Small models are viable: $7,800 training costs for competitive performance opens doors for indie developers.

4. Agent architecture matters: We're moving past naive agent implementations toward sophisticated patterns.

5. The discovery paradigm is shifting: AI search optimization may be the next frontier.