hadibuilds
AI Playbook 5 min read Updated 7 Oct 2026

How I run my company with 5 people and a lot of AI

The exact AI workflows we use at Heffl, department by department — code, sales, marketing, content, video and ops. Copy what fits.

Team
5
AI-written code
100%
Videos a day
4
01

Engineering

AI writes all of our code. Humans own the plan and the merge button.

Claude CodeAI code reviewGitHub

Feature → production

  1. Write the task in plain English: the problem, who it's for, what “done” looks like. Heffl
  2. Claude Code reads the codebase and proposes a plan. A developer approves it. Claude Code
  3. Claude Code builds it on a branch and runs the tests. Claude Code
  4. An AI reviewer checks every pull request for bugs and security issues. AI review
  5. A human reads the review, tests it and merges. It deploys automatically.

Steal this: Keep a CLAUDE.md file in your repo with your rules — stack, naming, things to never do. It's the cheapest way to make AI code look like your code.

02

Sales & CRM

Our AI CRM does the admin. We do the conversations.

Heffl CRMHeffl Agents

Lead → deal

  1. A lead comes in — website form, email or DM. Heffl CRM
  2. The AI agent creates or updates the record in the CRM. No manual data entry. Heffl Agents
  3. It reads the inbox and drafts replies and follow-ups. Heffl Agents
  4. Every morning it surfaces the next actions: who to follow up, which quote needs a nudge. Heffl Agents
  5. A human approves, sends, calls and closes.

Steal this: Let AI do the data entry, never the relationship. Approve every message before it goes out.

03

Marketing

Marketing isn't one person's job. Everyone on the team owns a number.

Market or DieGojiberry AI

Weekly marketing plan

  1. Pick a playbook — content, LinkedIn outbound, partnerships. Market or Die
  2. It turns the goal into a weekly plan with daily tasks for each person. Market or Die
  3. Everyone does their marketing tasks — not just “the marketer”.
  4. Friday scoreboard: hit your number, or your name goes on the wall.

LinkedIn outbound

  1. An AI agent finds people who match our ideal customer. Gojiberry AI
  2. The first message is short and human. No pitch, no product.
  3. When someone replies, a founder takes over the conversation.

Steal this: Small teams don't lack ideas, they lack consistency. Make marketing a weekly scoreboard, not a mood.

04

Content & LinkedIn

One real idea becomes five pieces of content. AI formats, I think.

Claude

Idea → posts

  1. Capture ideas from real work — a sales call, a bug, a lesson. Voice notes are enough.
  2. Claude turns the voice note into a LinkedIn post draft in my voice. Claude
  3. Same idea → carousel, Instagram caption (English + Malayalam) and a video script. Claude
  4. I edit until it sounds like me, then post.

Writing on LinkedIn with AI

  1. Ask for 10 hooks. Pick one. Rewrite it yourself. Claude
  2. One idea per post. Short lines. A real example with real numbers.
  3. End with a question, not a pitch.

Steal this: Give the AI your 10 best posts as examples and a list of words to never use. Generic AI writing comes from generic prompts.

05

Video

One editor plus AI ships four videos a day.

ClaudeCapCut

Record → every platform

  1. Script comes from the content pipeline. Claude
  2. Record a talking-head on a phone. One take, no fancy setup.
  3. AI edit: cut the silences, add subtitles and simple graphics. Claude skill
  4. AI thumbnail: pick a frame with eyes open, add the hook. Claude skill
  5. Final polish, then post to LinkedIn, Instagram and YouTube. CapCut

Steal this: Any edit you do twice, turn into a reusable AI skill. We did it for cuts, subtitles and thumbnails.

06

Distribution

We started distribution before the product was ready.

LinkedInInstagramYouTubeAppSumoEmail

How we get in front of people

  1. Share what you're building while you build it — not after.
  2. Launch early on a marketplace. We ran two AppSumo campaigns. AppSumo
  3. A weekly email to users, every week.
  4. Turn every customer question into a post.
  5. Comment funnels: “comment AI” → auto-DM → a page like this one.

Steal this: Distribution compounds slowly. Start it the same week you start the product, not the week you launch.

07

Operations

Docs, reports and research — Claude is our ops team.

ClaudeMCP

Ask → answer

  1. Proposals, SOPs and docs drafted from a voice note. Claude
  2. Reports pulled straight from the CRM: ask a question, get a dashboard. MCP
  3. Financial models and market research. Claude
  4. A human reviews and makes the call.

Steal this: Connect AI to your real data. An AI that can read your CRM answers questions a chatbot can only guess at.

Start here: copy this in one week

You don’t need all of this. Pick one department where you spend the most time and do this:

  1. Write down the workflow you repeat every week, step by step.
  2. Mark the steps that are typing, sorting or formatting. Those go to AI.
  3. Keep the steps that are judgement or relationships. Those stay human.
  4. Add one approval step before anything goes to a customer.
  5. Run it for a week, then turn the prompt into a reusable template or skill.

AI didn’t replace our team. It removed the work nobody on the team should have been doing.

The stack