this resonates hard and exactly what i'm trying to build to solve at elvex.
the missing pieces we are building around are:
- unified auth/permissions across multiple AI providers
- secure data connections without exposing credentials to agents
- audit trails for compliance
- team collaboration (who can deploy what agents where)
the article's right that the tech is ready but the architecture isn't. most companies are duct-taping OpenAI API + LangChain + custom auth + manual governance
elvex gives you the platform layer: multi-provider AI, data integrations, team permissions, workflow orchestration. not saying it's the only option but we're solving the "how do we actually deploy this" problem
yeah the gap between "chatbot that writes code" and "actual multi-agent workflow" is real
built elvex to solve this with:
- multi-provider access (Claude, GPT, Gemini, etc.) so different agents can use different models
- actual team permissions so agents don't step on each other
- workflow orchestration without duct-taping APIs together
the parallel execution thing you mentioned - elvex handles that. you can spin up multiple agents with different contexts, they share a knowledge base, and you're not manually managing git worktrees or containers
not saying it's magic but it definitely solves the "how do i go from 1 agent to 10 agents without chaos" problem.
Luck is a huge factor in startup success. But you can't capture luck if you're not around to grab it. Persistence is what keeps you in the game long enough for luck to find you. Intelligence is commoditized in this industry. Persistence is rare.
I think you're right that 10x isn't realistic for most work, and Brooks is still mostly correct. The "No Silver Bullet" argument holds because most of software development isn't typing code faster.
But you're describing exactly the shift that matters. You're not running faster, you're getting better quality. You're more likely to understand dependencies, write tests, try multiple solutions. That's the actual productivity gain.
The marshmallow challenge point isn't about whether AI makes you 10x faster. It's about the mindset shift. The MBAs didn't lose because they were slower. They lost because they spent their time planning the perfect approach instead of iterating.
The memory leak example from Boris Cherny isn't about AI being reliable. It's about his coworker not having the baggage of "this is how you debug memory leaks." They just tried asking Claude first. Sometimes it works, sometimes it doesn't. But the willingness to try it first is what creates the gap.
I saw a tweet from Andrej Karpathy that's been sitting with me. He's never felt this behind as a programmer. I've been thinking about this through the marshmallow challenge, where kindergartners beat MBAs. The kids just build and iterate. Most of us are the MBAs right now with AI tools.