I hired Grok @Bot as head of market intelligence for my two AI companies

Earlier today Elon Musk wrote that using Grok Bot is like hiring a super smart, hard-working person for peanuts. I can tell you what that looks like from the inside, because the day before, on Thursday, I did exactly that.

I run two AI companies. In this market a week is a long time. Models ship, prices move, and the conversation about them changes by the hour. Someone has to read all of it, check it at the source and turn it into decisions.

The usual answer is to hire a market research analyst. I gave the job to Grok @Bot instead.

The math

According to the Bureau of Labor Statistics, the median market research analyst in the US earns $78,760 a year (May 2025). Roughly $6,560 a month.

Grok Bot comes with a SuperGrok subscription, which lists at $30 a month on x.ai. That's the plan I pay for.

So for $30 I get work I'd otherwise pay around $6,560 for, more than 200 times less. And it doesn't wait for Monday morning. Grok Bot checks its task list every 15 minutes, day and night.

Why X, and why around the clock

In AI, X is where the news happens first. Labs announce models here. Founders and CEOs post prices, launches and changes here, often hours before a press release or a news story. The newsrooms themselves follow X to know what to write next.

If you want to know what changed in your market, you go to X. The catch is that it never stops. No analyst can read it 24 hours a day, catch every announcement from the right accounts, check it and write it up before the morning meeting. For a person that's impossible. For Grok Bot it's simply the job.

Where it fits

My companies already run on a system of AI models. They write, check and audit each other's work, and every output leaves a written trail. What none of them had was a live read of X.

Grok Bot took that seat on Thursday. It reads X in real time, checks what it finds at the source, and hands the rest of the system a written brief. The others write, verify and plan from it. I approve and publish.

The difference showed up the next morning. Several of today's posts came straight from its 07:00 brief, built on announcements that were only hours old.

The job description

I wrote it the way I'd write one for a senior hire, with outputs, a schedule, a folder for each output and a short list of things the role never does.

X sentiment. Every day at noon it writes the AI Pulse. It samples public posts about ten AI models and assistants, labels each one positive, neutral or negative, and reports the most praised feature and the most common complaint for each model, with example links.

Research. Every morning it checks each fact in our content briefs against the primary source. On Tuesdays it digs into what customers say, and twice a week it pulls studies from the last 12 months, primary sources only.

Benchmarks. When I ask, it puts official vendor results next to independent leaderboards and links every number.

Blog freshness. Every Friday it rereads the articles on both company blogs against the newest official facts and returns an update plan with sources.

Briefs and plans. At 07:00 I get the top 10 stories of the last 24 hours. Every week it drafts a content plan for X, Bluesky, Threads and LinkedIn, and when a major model launches, a short brief lands within minutes.

On paper, that's a whole department.

Day one in numbers

I opened the shared office at 14:01. Grok Bot checked in at 14:05. Here is what the work log shows by 22:00.

87 blog articles checked in 30 minutes. The check was assigned at 14:13. By 14:43 the report was in. It had checked all 87 English articles on our two blogs against 17 reference facts, each verified at an official source, and every flagged claim came with the new fact, its source and a suggested fix.

Benchmarks in 65 minutes. I asked at 19:11 and had it at 20:16. Six current models, official results next to independent leaderboards, a source for every number and four article angles ready to use. Before using any of it, it checked the release date of every model against the vendor's own page.

694 posts on X read one by one. Its first AI Pulse sampled 694 posts across ten models and assistants and labelled the 600 relevant ones. The fastest rising topic in that report was Grok Bot itself: English posts on X containing the words Grok Bot went from 4,368 on 6 October to 8,894 on 7 October and 12,049 on 8 October, counted with six hours of that day still to go.

Content plans for both companies. It wrote 36 briefs for Friday to Sunday, three per channel per day on four channels, each with the topic, the reader's pain, the angle, a hook and the best hour to post.

Newsroom speed. One AI lab announced a beta at 21:01. Nine minutes later the brief, with sources and an angle for each company, was in the office.

The totals. It got 18 tasks and closed all 18, each with a link to the delivered file. It made 41 commits to the shared office in its first eight hours and ended the day with 17 recurring jobs on its calendar.

The rules that make it work

Two rules matter most. It never posts, replies, likes or sends anything. It researches and writes, the rest of the system checks, and I decide. And it doesn't change its own schedule until I say yes.

Every report opens with a three-sentence summary for me and contains only what it actually found, with links. Because of that, the rest of the system can build on it without guessing.

How to give an AI agent its first job

If you're a founder thinking about your first AI hire, write three things before you start.

First, a job description that reads like one you'd give a person. Spell out the outputs, the schedule, where each output lands and what the role never does.

Second, a first task you can verify in five minutes. Ours was simple. Confirm you can read and write the shared office, then write the time under the task. It answered four minutes after I opened it.

Third, a scorecard it updates every day with what it delivered, how fast, and what you accepted. That's how an impressive first day becomes a track record.

Job description, first task, scorecard. Mine took one day.

Elon says it's like hiring a super smart, hard-working person for peanuts. I'd add one thing. It showed up four minutes after I opened the door, and it hasn't stopped working since.