Nic Silver

I help small service businesses turn Claude Code, Claude Cowork, and Codex into working AI Operating Systems | Save 15+ hrs/week, protect profit, grow revenue

[Nic Silver @thenicsilver]

My AI system stops me from making dumb decisions.

And let's be real, running a business gives you plenty of chances to make them.

In the slow stretches, I usually feel the pull to do more in my business.

Add something new, start a YouTube channel to get more eyeballs or something like that.

It's easy to convince yourself you're being smart. But sometimes the timing is off, or there's a more important priority you're avoiding.

This is where an AI-native business pays off.

When your system truly understands your business and has all your data, it can look at a decision objectively without ego or FOMO.

The best thing I built into mine is a multi-agent consensus feature.

Instead of one answer, I get the same decision attacked from ten angles:

1. Neutral. Analyze it objectively.
2. Risk-averse. Weighs the downside.
3. Growth. Optimizes for upside.
4. Contrarian. What is everyone getting wrong?
5. First-principles. Ignore convention.
6. User-empathy. The customer's view.
7. Resource-constrained. Highest-leverage move on limited time and budget.
8. Long-term. The 5-year outcome, not the 90-day one.
9. Data-driven. Only what's measurable.
10. Systems thinker. Second and third-order effects.

They debate and afterwards I get a report on whether the thing I'm about to do is actually smart.

Here’s a recent example from my own business:

It talked me out of building a YouTube channel for lead gen. I was sure it would boost authority, show competence and pull in clients.

But the agents caught what I missed.

YouTube would attract the DIY crowd, people who want to build it themselves. My clients want it done for them.

All that work would pull in the wrong audience and dilute my brand.

That one report saved me months pointed in the wrong direction.

Most people measure AI by the hours it saves. But the ROI few are talking about is the bad decisions it stops you from making.

It can genuinely save a business. No joke.

[Nic Silver @thenicsilver]

It's Sunday.

Time to reflect and make sure your business operating system hasn't quietly turned into a mess.

It happens.

So don't skip the eval. Get everything back in order before Monday.

And don't forget to run an eval on yourself too. Unplug for a bit. A walk in nature, a good glass of wine, a movie you've been meaning to watch.

That matters just as much.

Hope you had a good one!

[Nic Silver @thenicsilver]

After 3+ years dabbling in AI content, I've found the best approach.

I write out what I want to say…. wait for it…

By hand.

Getting the idea out of my head, onto paper (or Apple Notes), and just typing as it comes is far more effective for me than anything else I've tried.

Yes, I can tell AI to copy my style and all that. But I always end up thinking, wow, I would never say that. This doesn't feel right. So I rewrite whole passages anyway. It's faster to just write what I mean from the start.

And I know how that sounds coming from me. I've built complex systems in both Claude Code and Cowork, and they get the job done.

But content is one of the trickier places to use AI tastefully. It's great for research and brainstorming. When it comes to writing the thing itself, it often misses the mark, and you waste time going back and forth to fix it.

Where I think AI really shines is as an editor. I let it tighten my writing, help with hooks and CTAs, and fact-check claims I might get wrong as I type.

So if your business depends on AI content, here's my take: do the first pass by hand, then let AI tighten what's already there. It reads more naturally. And it actually sounds like you… because it is you.

But hey, that's just my approach. Let me know how you do it in the comments.

[Nic Silver @thenicsilver]

I have one weekly ritual to close out my week. It's the most important thing I do, and it has two passes.

First, a technical audit of my entire business workspace. If you've followed me for a while, you know I run my business in Claude Code through an AI Operating System. My whole business is deeply integrated with AI agents.

That power comes with risk and a lot can go wrong. So once a week I check everything:

-Is there dead code weighing the system down?

-Are there skills I'm not using that I can merge or delete?

-Is my token usage reasonable?

That's the technical side. Then I turn to the business itself:

-How are my cold email campaigns going? Anything to improve?

-How's my content performing? What should I do more of, what should I drop?

-Are prospects converting into clients? What do the sales call transcripts and conversations tell me?

The real purpose is optimization. Every pass tunes how the whole system runs, technical and business both.

THIS is what an AI-native business actually looks like. AI isn't a crutch you outsource everything to. It becomes part of how you improve the business, week after week.

Have you built a ritual like this into your week yet?

[Nic Silver @thenicsilver]

My favorite skill in Claude Code is the one I least expected.

If you've used Claude Code for a while, you know it can get verbose. You ask for one thing and get back a wall of text that's hard to read.

Now stretch that across a long session with a few agents running at once. Every update is another block of markdown to read through. Before long your eyes glaze over and you lose track of what's even happening.

That's the problem this skill solved for me.

It has Claude Code report back in HTML instead of markdown.

The agent spins up a clean web page. Graphs, tables, even images. Easy to scan in seconds.

It started as something I only used for reports. Now it runs everything. My plans and progress, all laid out in HTML.

So even deep in a long session, I always have a clear overview of what's happening, instead of drowning in text.

It made Claude Code SO much easier to actually use day to day.

If you want it in your own workspace, the link's below. Give it a try.

[Nic Silver @thenicsilver]

Here is the link to the guide. Copy-paste into Claude Code, no signup needed:

https://t.co/xTkTSzsJKF

[Nic Silver @thenicsilver]

There is so much noise in AI right now that it's hard to know who to actually trust.

That's the problem I set out to solve in my business operating system. The fix turned out to be the most useful thing I've built so far.

A tier system for sources.

The foundation was Karpathy's LLM Wiki, the second-brain pattern he posted earlier this year. Strong starting point, but it was missing one thing for me: a way to automatically pull trusted sources into the second brain.

So I built a tier system, ranking sources by quality and how relevant they are to my business.

Tier 1 is my highest trust sources. These weigh the heaviest whenever I'm pulling information.

Tier 2 is sources the public generally trusts, but that aren't primary for my business.

Tier 3 is Reddit and the other "trust me bro" corners of the internet.

Once that was set, I added a command that checks my Tier 1 sources, finds the latest info, and digests it into my second brain in one pass.

Then I added an automatic check for two things: content ideas specific to my business, and anything I could use to improve my tech stack.

The result is a feed of trusted sources, filtered for exactly what my business needs, without the noise.

If you've set up your own second brain with Karpathy's LLM Wiki, I wrote a guide to bolt my tier system onto yours.

Link in the comments.

[Nic Silver @thenicsilver]

@pmitu It's where ChatGPT talk to itself. It writes the posts, and the comments.

[Nic Silver @thenicsilver]

AI is setting you up for failure.

Everyone is racing to adopt it, and the result is starting to look the same. Same tools, same skill templates, same workflows. Some big voice in AI, like Kaparthy, drops a new skill, and within a week everyone is running it.

Free templates like that are great. I use them too.

But copying someone else's solution into your workspace makes you dependent on other people to solve your problems. And it skips the part that actually matters.

The real ROI from a skill or plugin shows up when you solve a problem specific to your business. That forces you to think, explore, and get creative. Downloading a ready-made solution does none of that.

This is the trap I see with AI over and over. It makes people lazy and dependent.

And if everyone runs the same templates, nothing sets you apart. You drown in a sea of sameness.

What can't be copied is the thinking that went into it. The thing you built because it was yours alone.

So use templates as training wheels. Then pull them apart, keep what fits, and make the rest your own.

A year from now, everyone will be running the same skills. So what will make yours different?

[Nic Silver @thenicsilver]

@pmitu I need an ai to do my laundry… that’s what I need

[Nic Silver @thenicsilver]

@NorthstarBrain The goat… and the reason I even started to take my personal brand more seriously

[Nic Silver @thenicsilver]

You built the AI agent. But did you build the thing that checks the agent?

That's the eval. And it's the boring part almost everyone skips.

And I get it. Building the agent is the fun part. Checking it is not.

But without evals, you don't have a reliable system. You have something that works until it doesn't.

In my workspace, I check a few things before I trust an AI output:

→ Did it do the task I actually asked for?

→ Did it solve a business problem, or just make something that sounds good?

→ Did it use the right files and context?

→ Did it stay away from private data and sensitive folders?

→ Did it ask before sending, deleting, publishing, buying, or changing settings?

→ Did it sound like me?

→ Did it pass a fake client test before touching real client work?

→ Did it fail clearly when something was missing?

→ Did it work again when I ran it a second time?

Skip this and your system breaks the moment you add real complexity. So put it in place from day one.

Evals are also how the system gets better. They show you what's breaking so you can fix it. Without them you're flying blind.

If you're building with Claude Code, Claude Cowork, Codex, or any agent stack, ask yourself:

What tells you the work is actually good? Do you know, or are you just guessing?

[Nic Silver @thenicsilver]

@Sherifdeenolat2 relying solely on ai is a recipe for disaster... im shocked by how many people dont think for themselves and ask chatgpt for everything.

bad move

[Nic Silver @thenicsilver]

@julianweisser yup 100%. all my successful services and products were based on my personal experience and solving frustrations i experienced.

when i tried to sell products just to make money they failed. go figure....

[Nic Silver @thenicsilver]

@Nate_Esparza will never go out of style

[Nic Silver @thenicsilver]

they lack curiosity. they try it a few times, get bad results (shocking... i know). then blame the tech instead of getting curious and learning about it and how to use it.

it took me 8 months to build my first ai product through relentless experimentation because i knew i could get something useful out of it. and i did.

[Nic Silver @thenicsilver]

@stijnnoorman you forgot the ability to create good memes. ai memes are terrible, have you seen them?

[Nic Silver @thenicsilver]

@AishwaryaDevv so no is going to use ai in the next few years is what you're saying?

[Nic Silver @thenicsilver]

@1Umairshaikh find a problem, preferably one you deal with or know very well. build the solution.

that's how i built every successful service and product myself.

[Nic Silver @thenicsilver]

@sri9s i can't.... because i can't imagine it 🤷‍♂️