Claude Code & AI Agents in Practice: 5 to 100 Commits a Day | Tejesh Vaish (IIT Kanpur, Lead AI Engineer)
Notes from my unfiltered conversation with Tejesh Vaish, Lead AI Engineer in Dubai and IIT Kanpur Computer Science graduate, on AI coding agents, Claude Code skills and getting real work done with AI.
Watch the full episode
Most people use AI the same way they did in 2023: ask a chatbot a question, copy the answer, paste it somewhere, fix the error, paste the error back. It helps, but it doesn't scale.
A small group of engineers and founders work very differently. They run several AI agents at once, hand repeated work to reusable "skills", and spend their own time reviewing and deciding instead of typing.
Tejesh Vaish is one of them. He studied Computer Science at IIT Kanpur, worked at Abacus.AI and Simbian, and is now a Lead AI Engineer in Dubai, building AI agents for e-commerce. Alongside his full-time job he built Homeo-AI.com, an app used by homeopathic doctors in many countries.
On the Codersarts Podcast, he showed exactly how he works, with live demos. Here are the ten lessons I took away, and what each one means if you want AI agents working inside your own business.
1. A chatbot answers. An agent works in a loop.
Tejesh's explanation is the clearest I've heard. With a chatbot, you are the loop: you ask, copy, run, see an error, paste it back, ask again.
An AI agent runs that loop on its own. It decides the next step, calls tools, reads the result, and either gives you a final answer or keeps going until the task is done.
"If you program this loop, then it becomes an agent, because now the agent is deciding what to do next."
What it means for your business: The value of AI isn't in better answers. It's in removing the manual copy-paste steps between answers. Any process where a person keeps moving information between tools is a candidate for an agent.
2. His output went from ~5 commits a day to ~100
Tejesh shared his GitHub activity live. In 2024 his graph showed a handful of contributions a day. In 2026 it shows close to 100 on busy days.
He's clear that this didn't happen overnight. It came from months of building workflows the AI could run reliably.
What it means for your business: The productivity jump from AI is real, but it comes from setting up workflows, not from a subscription alone.
3. Skills are cached solutions to problems you've already solved
Tejesh's public repo has over 130 Claude Code skills, and he now uses around 300 day to day. He doesn't sit down and plan skills. They come from real work:
Do a task with the AI
Notice it's complex or repeated enough to keep
Package the instructions, references and code into a skill
Next time, call the skill instead of prompting from scratch
He compares it to dynamic programming: solve a sub-problem once, cache the answer, reuse it.
A skill can be as simple as a saved set of instructions, or as advanced as prompts plus code files, reference docs and scheduled jobs.
What it means for your business: Every repeated process your team runs, such as reports, research, follow-ups or proposal drafts, can become a skill. Once captured, it runs the same way every time, whoever triggers it.
4. Start with one simple, real automation
One of Tejesh's most useful skills took him a few minutes to create. His company kept losing track of customer messages, and replies were slow.
The fix: a skill that runs every hour, checks all email and Slack threads, finds customer messages with no reply in the last three hours, and posts a report to the team channel.
It needed only the Slack and email connections (MCP) and a plain-English description of the task. The result is that no customer message goes unanswered for long, and the team stays accountable.
What it means for your business: The best first automation is small, painful and frequent. Follow-ups, lead alerts, daily summaries and status reports are good places to start.
5. Do everything inside your AI agent, even when it feels slower
Tejesh's main tip for anyone starting out: force yourself to do your work from inside Claude Code or Codex.
For the first two or three weeks it will feel slower, because you are reviewing output and building skills. He compares it to starting exercise: the hard part comes first and the results come later. But it shows you quickly what the AI does well (automate it) and what it doesn't (keep it human, or find a better approach).
What it means for your business: AI adoption has a learning curve. Teams that push through the first month gain a lasting advantage over teams that give up when it feels awkward.
6. Run agents in parallel and review like a pilot
During the demo, Tejesh had one agent analyse a GitHub repo and draw an architecture diagram while a second agent read his inbox and drafted a thank-you email. Day to day, he actively runs up to ten agents, with more skills running in the background on schedules.
His analogy: a pilot spends most of a flight monitoring, not flying. They step in only when something goes wrong.
"Everyone has to aim to become the pilot of their life through these agents."
What it means for your business: The role of skilled people shifts from doing every task to orchestrating and reviewing. That only works when the agents are reliable, which is why the setup matters.
7. Ask AI to explain with diagrams, not paragraphs
AI tends to produce long walls of text. Tejesh asks for diagrams instead. To understand a new codebase, he gives Claude Code the repo and asks it to explain and visualise it, then asks follow-ups on any part that's unclear.
What it means for your business: For onboarding, documentation and handovers, visual explanations generated from your real codebase save hours of reading.
8. Security: be careful with credentials and third-party skills
Tejesh raised three risks every team should know about:
Data settings: check your AI tool's data-sharing and training settings and turn off anything you don't want shared
API keys: never paste API keys into your agent; give it tools that make the calls instead
Third-party skills: a skill copied from the internet can contain malicious code that steals credentials, so scan or review every skill before installing it
We have seen the cost of this ourselves: a leaked environment file can drain AI API credits within hours.
For business inboxes and sensitive systems, a custom connector (for example, your own Gmail API app or custom MCP server) gives you more control and visibility than connecting everything directly.
What it means for your business: AI agents need the same security thinking as any other system with access to your data: least privilege, secret management and review.
9. Building is no longer the bottleneck. Knowing what to build is.
Tejesh built Homeo AI (https://homeo-ai.com) on top of an open-source interface, adding RAG pipelines, semantic and keyword search, and agent workflows. He tested it with doctors in Lucknow, marketed it through WhatsApp, Meta ads, YouTube, webinars and clinic visits, and now has paying users across many countries.
His view on what matters now:
"Building has not become the bottleneck. Realising what to build, who the audience is and how you will reach them has become the bottleneck."
He also gave a warning many founders need: before building an app, ask whether it still needs to exist, or whether users could do the same thing with a general AI assistant and the right tools.
What it means for your business: Spend more time on problem discovery, users and distribution. Use AI and experienced engineers to compress the build.
10. Fundamentals still decide who wins
Tejesh is clear that AI makes strong fundamentals more valuable, not less:
Know your domain deeply, whether that's DSA and system design, sales concepts like TAM and ICP, or DevOps tools like NGINX and Cloudflare
Treat AI like an employee you have hired: you need enough expertise to judge its work
For students and job seekers, academics, DSA and medium-level coding problems still matter, because they are the clearest signal of ability in a short interview
His cost advice is practical too. He spends heavily on AI himself, but recommends everyone start with at least an entry-level paid plan, and check student, open-source and startup programmes for free credits.
What it means for your business: AI amplifies expertise. Pair AI agents with people who understand the domain, and the output improves quickly.
Where to start: an AI agent readiness checklist
Step | Question | Good starting point |
Pick a process | What repeated task wastes the most time? | Follow-ups, reports, research, lead alerts |
Connect data | Which tools does the task touch? | Email, Slack, CRM, docs (via MCP connectors) |
Capture it | Can the steps be written down? | Turn the instructions into a reusable skill |
Schedule it | Does it need to run on its own? | Hourly or daily scheduled runs |
Secure it | What can the agent access? | Least privilege, no raw API keys, reviewed skills |
Review it | Who checks the output? | Human review until it is consistently right |
Build AI agents and automations with Codersarts
Not every team has the time to build and maintain its own agent setup. At Codersarts, we help founders, startups and businesses put AI agents to work on real processes, securely and reliably.
We can help you:
Build custom AI agents for support, sales, operations, research and internal tools
Create Claude Code skills and workflows tailored to your team's processes
Build custom MCP servers and connectors for your email, CRM, databases and internal systems
Build AI-powered apps and MVPs with RAG, search and agent workflows, like Homeo AI
Secure your AI setup with proper credential handling, access control and skill review
Want AI agents handling your repeated work? Tell us what you'd like to automate: https://www.codersarts.com/contact
Explore our AI services: https://ai.codersarts.com
Connect with Tejesh Vaish
LinkedIn: https://www.linkedin.com/in/tejeshvaish/
Website: https://www.tejeshvaish.com/
Homeo AI: https://homeo-ai.com
Tejesh shares blogs on his website and is writing a book on working with AI coding agents. He welcomes questions on LinkedIn, so tag him if you build something inspired by this episode.
Frequently asked questions
What is the difference between an AI chatbot and an AI agent?
A chatbot answers one question and stops, so a person has to move information between steps. An AI agent runs in a loop: it decides the next action, uses tools, checks results and keeps working until the task is complete.
What are Claude Code skills?
Skills are reusable packages of instructions, reference documents and sometimes code that tell Claude Code how to perform a specific task. Once created, a skill lets you run the same workflow repeatedly without prompting from scratch.
How do I create my first Claude Code skill?
Do a real task with Claude Code first. If it's complex or repeated, ask Claude Code to turn what you just did into a skill, including the steps, rules and resources it should use next time.
Is it safe to install skills from GitHub?
Only after reviewing them. A third-party skill can contain malicious commands or URLs that steal credentials, so read or scan any skill before installing it, and never paste raw API keys into your agent.
How much does it cost to use AI coding agents?
Costs range from entry-level plans of around $20 a month to much higher for heavy users. Student, open-source and startup programmes may offer free or discounted access.
Can Codersarts build AI agents for my business?
Yes. Codersarts builds custom AI agents, automations, MCP connectors and AI-powered apps. Share your requirements at https://www.codersarts.com/contact.
Watch the full conversation with Tejesh Vaish on the Codersarts Podcast: https://youtu.be/OB99jxLKkkk
Want to be a guest on the Codersarts Podcast? Email jitendra@codersarts.com with your name, role and the topic you'd like to discuss.
Comments