The network for creativity
Join 1.25M professional creatives like you
Connect with clients, get discovered, and run your business 100% commission-free
Creatives on Contra have earned over $150M and we are just getting started
Back to feedPost
Lately I've been getting some really weird questions from clients. Weird enough that I'm about 99% sure they've come from AI.
One client asked me what date and time the ads went live. time??????? What are you gonna do with that information? 😭
Another one asked whether the budget for the new campaigns was a net new budget or taken out of existing campaigns.
...it's literally their first meta campaign. There are no existing campaigns.
I genuinely don't know what people are asking AI and now I have to spend the next 20 minutes answering questions that don't even matter.
At some point we need to stop outsourcing common sense and stop treating AI like the final authority.
It's starting to replace thinking and this is really scary.
The network for creativity
Join 1.25M professional creatives like you
Connect with clients, get discovered, and run your business 100% commission-free
Creatives on Contra have earned over $150M and we are just getting started
Related posts
700 followers on Contra. 🫶
A small number, but a big milestone for me.
700 → 1K. Let’s keep going. ⚡
PS: Yes, the post is AI-generated.
At this point, I’m just supervising. 😂
A personal identity experiment for Super Intelligence.
The symbol combines a star and the letters SI into one compact mark.
Logo design + motion.
Excellent use of geometry in the logomark!
In 90 days, AI stopped being a tool I use and became a teammate I manage.
Here's what actually changed, from where I sit as a software engineer:
→ Agents went mainstream. JetBrains found 90% of professional developers use coding agents weekly and 68% daily. Claude Code usage doubled since January (18% → 39%). Codex went 3% → 16%.
→ Models learned to drive computers. GPT-6 Astra (Sep 3) was pitched on computer use and software engineering, and shipped with gated access over its cyber capabilities.
→ Output limits blew open. Gemini 4 Argon (Sep 30) can write up to 1M tokens in one response, up from 64K. Also gated at launch.
→ Agents did real, checkable work. Dozens of Claude agents formalized Fermat's Last Theorem in Lean in 11 days: 13M lines, machine-verified. (Formalization, not a new proof. The verification is the point.)
→ The plumbing grew up. MCP's July spec went stateless, so agent tooling now deploys like ordinary web infrastructure.
→ Open weights kept pace. Alibaba released Qwen3.8-Max as a 2.4T-parameter open-weights model.
My take: code generation is getting cheap. Judgment isn't.
Anthropic's agentic coding report says engineers use AI in roughly 60% of their work but can fully delegate only 0–20% of tasks. The bottleneck has moved to architecture, review, testing, security, and keeping things alive in production.
That's where I spend my time: shipping AI features (agents, RAG, LLM integrations) into real products, with the guardrails that make them survive contact with users.
What's one task you'd hand to an agent tomorrow, and one you never would?
#AI #SoftwareEngineering #AIAgents #LLM #FullStack
managing agents instead of just using them is exactly the shift I've felt too ngl