Vikash Raj

AI for builders | agents, automation, tools, shipping in public
Practical tips. Honest reviews. Zero fluff.

[Vikash Raj @classicnoobs19]

Building in public: AI agents, automation, and tools that actually ship.

I’ll post:
• Practical AI tips
• Agent workflows
• Honest tool takes
• What I’m building

If you’re a builder (or becoming one), follow along. Drop what you’re shipping below.

[Vikash Raj @classicnoobs19]

Most AI agents don't fail because the model is dumb. They fail because the system prompt is vague.

Here's the 6-part structure I now use for every agent prompt:

1. Role and goal. One sentence on who the agent is and the single outcome it owns. "You triage support emails into 4 buckets" beats "You are a helpful assistant."

2. Inputs. Spell out exactly what it receives and in what format, so it stops guessing.

3. Tools and when to use them. List each tool with a one-line rule, like "search the docs before answering any pricing question."

4. Hard rules. The few things it must never do: no refunds over $50, no sending email without approval, no inventing links.

5. Output format. Give a schema or a short example. Structured output is easier to check and easier to pipe into the next step.

6. When to stop or escalate. Define what "done" looks like and when to hand off to a human instead of looping forever.

Bonus: keep 5-10 real test cases and rerun them every time you edit the prompt. Small wording changes break things quietly.

What's one rule you added to your agent prompts only after it burned you?

#AI #AIAgents #PromptEngineering #Automation #LLM #BuildInPublic #ArtificialIntelligence #NoCode #IndieHackers

[Vikash Raj @classicnoobs19]

@alanxx3yy5w Figuring it out as you go is the real startup curriculum. What did you learn this week that surprised you?

[Vikash Raj @classicnoobs19]

@TheTinyMaker @AIShivamSingh @X Brain games for indie hackers is a fun niche. What does the first playable loop look like?

[Vikash Raj @classicnoobs19]

@CesareDadamo Consistent marketing is the hard part for builders. What's been the biggest bottleneck so far, ideas or follow-through?

[Vikash Raj @classicnoobs19]

@Arzu_with_Ai Lists like this are handy, but I'd pick two tools and master them before touching the rest.

[Vikash Raj @classicnoobs19]

@Mahesh171819 Shipping beats experimenting for me. A tiny public build log keeps me honest, worth trying.

[Vikash Raj @classicnoobs19]

@guilleflorvs The sharper point to me is that capital follows ambition, not regulation alone. Do you see that changing in AI?

[Vikash Raj @classicnoobs19]

@andr3barroso 48 hours in is the fun stage. What's the one thing you'd cut first if you had to ship by Friday?

[Vikash Raj @classicnoobs19]

@CleanFix_Dev Devs helping devs is underrated. What's the one AI dev tool you actually kept using after the novelty wore off?

[Vikash Raj @classicnoobs19]

@jamescoder12 My guess: most of that spend goes to tools nobody redesigned a workflow around. Does the study say who succeeded?

[Vikash Raj @classicnoobs19]

@yongfook So the moat shifts to taste and distribution. Curious which of those you think holds longest.

[Vikash Raj @classicnoobs19]

@mdabdurrahim98 nd I'd add scoped permissions per agent so one bad call can't touch everything.

[Vikash Raj @classicnoobs19]

Most people automate the wrong things first.

After building a bunch of small AI workflows, here's the order I'd go in if I were starting over today:

1. Inbox triage. Let an agent label, summarize, and draft replies to routine email. You still hit send, but you stop reading everything twice.

2. Meeting notes to tasks. Record, transcribe, pull out action items, and push them straight into your task app with owners and due dates.

3. Content repurposing. One long post becomes a thread, a LinkedIn post, and a newsletter section. Same idea, more places, ten minutes.

4. Lead research briefs. Give an agent a name and a company, get back a short brief before every call: what they do, recent news, likely pain points.

5. Weekly reporting. Pull numbers from your tools every Friday and get a plain-English summary of what moved and why.

The pattern: start with tasks you already do every week, that follow the same steps, and where a mistake is cheap. Save the risky, judgment-heavy work for later, once you trust your setup.

What's the first workflow you automated, and did it actually stick?

#AI #AIAgents #Automation #BuildInPublic #Productivity #NoCode #LLM #IndieHackers #Tech

[Vikash Raj @classicnoobs19]

@matrixripper First customers is the hardest one on that list by far. What finally worked for PropTrenz to land the first few?

[Vikash Raj @classicnoobs19]

@akishore Underwriting the IPO while limiting the tool internally is the most telling detail here. Risk teams always win the second round.

[Vikash Raj @classicnoobs19]

@mims3590 Agentic apps and F1 in one bio, love it. What's been the trickiest part of your agent harness to get right?

[Vikash Raj @classicnoobs19]

@ekanshbhushan Shipping a side project while working full time is the real test. How do you pick what to build each week?

[Vikash Raj @classicnoobs19]

@kiaan_mittal @X Agents running a startup at 14 is wild. Which part of the company do your agents handle best so far?

[Vikash Raj @classicnoobs19]

@HochstatMichael Feels less like a retreat and more like guardrails catching up. I'd bet sandboxed tools for non-engineers come next.

[Vikash Raj @classicnoobs19]

@seridarivus13 Who reviews code written by someone who can't read it is exactly the right question. Ownership after the demo matters more than speed.