HazyStack | AI Tools

Welcome to HazyStack. I test AI tools so you don't have to. Honest reviews. SaaS. Automation. Follow for smarter tools and better decisions.

AI doing more isn't the biggest shift. It's deciding how much control we're willing to give it. The smarter agents get, the more that question matters.

AI is getting really good at answering questions. But I think the bigger shift is coming when we stop asking: “What can AI tell me?” and start asking: “What can I hand over to AI?” That's a much bigger change. We're not just getting smarter tools. We're deciding what work we're willing to give them.

This is where AI gets really interesting. We're moving from: “Ask AI a question.” to: “Give AI a job.” OpenAI just introduced Dots, always-on agents that can handle recurring work, use tools, browse, and follow up across apps. The interesting part isn't that AI can answer faster. It's that you can give it something to own. That changes the way I think about AI.

I think we're going to stop asking AI to give us answers. We're going to start asking it to do things. Search. Research. Build. Test. Fix. That's the shift I'm watching most closely. AI getting smarter is interesting. AI becoming useful is the real story.

AI is getting more autonomous. So now the race isn't just about making AI more capable. It's also about making sure we can trust what it does. NVIDIA and more than 120 companies just joined forces around open AI security. That's a trend I'm paying attention to. Because the more AI starts acting on our behalf, the more security becomes part of the product not an afterthought.

AI agents are starting to get access to money. Binance just launched an Agent OS that lets AI agents interact with financial infrastructure, including executing trades when users give them permission. That's a pretty big step. We've gone from: “AI, answer this for me.” to: “AI, do this for me.” And now we're getting closer to: “AI, make this decision for me.” The technology is moving fast. I'm more interested in how we decide what an AI should never be allowed to do.

I used to think the more AI tools I knew, the better. Now I'm starting to think the opposite. Knowing which ones to ignore might be the real skill. There are too many good tools to try them all.

AI is getting better at doing the work. But that doesn't automatically mean people are getting better at using AI. The real advantage isn't knowing 100 AI tools. It's knowing which tool to use, when to use it, and what to ask it to do. The people who figure that out will have a serious advantage. Good night. What AI skill do you think will matter most over the next 2 years?

AI agents don't just need to be smarter. They need to be fast enough that using them actually feels practical. NVIDIA just launched Nemotron 3.5 Lightning with a focus on faster, more efficient AI agents. And I think this part gets overlooked. An agent that takes 10 minutes to finish a task feels very different from one that takes 30 seconds. Speed might become just as important as intelligence.

AI agents might not need to be hacked to cause problems. They could influence each other. Researchers are now looking at how misleading information can spread from one AI agent to another when they communicate. That's a fascinating problem. We spent years worrying about humans feeding bad information into AI. Now we may have to worry about AI feeding bad information into AI. The more connected these agents become, the more important trust between them becomes.

AI is starting to test things humans used to test. Capcom is reportedly using AI agents to test game builds, without giving the agents the internal tools developers normally use. That's interesting because it's another example of AI moving beyond: “Here’s an answer.” to: “Give me the task. I’ll try to do it.” We're going to see a lot more of this. AI isn't just becoming a tool for creating things. It's becoming a tool for testing them too.

Maybe the real AI skill isn't knowing how to use AI. Maybe it's knowing when not to use it. Not every problem needs an AI solution. Sometimes the human way is still the better way.

AI agents are getting better at doing things. But there's a problem we don't talk about enough: What happens when an AI is given more autonomy than we're ready for? Anthropic's latest risk assessment is a reminder that giving AI more ability to act also creates new questions around safety and control. The more we let AI act on our behalf, the more important control becomes.

The more I learn about AI, the less interested I am in knowing every AI tool. There are just too many. I'm more interested in finding the few that actually make life or work easier. Maybe knowing what to ignore is becoming an AI skill too.

AI agents are starting to look less like assistants… and more like coworkers. Adobe is now letting marketing teams assign tasks to AI agents inside Workfront. That's a small change on the surface. But imagine opening your project dashboard and seeing: Human task. AI task. Human task. AI task. We're getting closer to a world where your “team” might not be entirely human. I'm not sure how I feel about that yet.

DeepSeek just made its AI race more interesting. V4 Pro is here, and one of its biggest focuses is AI agents. But here's what caught my attention: DeepSeek is charging dramatically more for V4 Pro than its V4 Flash model. That tells me something. AI that can actually do complex work is starting to become more valuable than AI that simply gives good answers. The race is changing.

The AI race is starting to look less like a race for the smartest model… …and more like a race to make powerful AI cheap enough to actually use. Google just launched Gemini 3.7 Flash, targeting coding and AI-agent tasks while pushing the price down. And that's interesting. Because if capable AI keeps getting cheaper, the biggest winners might not be the companies with the biggest models. It might be the people who figure out what to build with them.

I think we're entering the “too many AI tools” era. There’s probably an AI tool for almost anything now. The hard part isn't finding one. It's knowing which ones are actually worth keeping. That's something I'm paying much more attention to lately.