Bootstrapping a startup is hard. Security space is crowded. LLM agents are finicky. AI is expensive.
Yet, it's been relatively easy for us at Transilience AI to get paying customers and delightful users while keeping our costs minimal. To put it in perspective, we have one frontend engineer: Muzaffar Hossain, one backend engineer, Alessio Mauro, one designer, Garima Sadhnani who worked on the product full time starting in June.
Our AI stack costs are minimal, enough to be covered by our revenues.
If we can do it, anyone can do it. No need to raise millions of seed money. Some lessons learnt so fellow entrepreneurs can jump in into starting their own companies.
Hone in on the use case and know your architecture
We can boil the ocean with LLMs, but you should not try to LLMs on all use cases. As you are talking to customers, see if the pain point and use case that customer is expressing is only solved by LLMs.
For example - tell me if this vendor has a workaround for a given CVE in their advisory. It cannot be solved programmatically with out LLMs.
Know your architecture leverage
There are core capabilities of LLM and agentic AI architecture, if you get them right, you can solve several use cases using the same architecture components.
For example - structured output extraction from several different formats of information. If you get that component right, you can parse out threat intel advisories from CISA or exploit code from metasploit. RAG , if you get RAG right, you can solve compliance use cases or vendor documentation parse out use cases.
Pick the right team
AI engineering expertise can be expensive, but can be worked around.
For example - You only need one LLM expert (who is the most expensive). The rest can be done by good python and react engineers. As the team adjusts to LLMs , smart engineers would develop the taste for LLMs. Our front end team joined with out much LLM expertise but now they are well versed with AI UX patterns.
Pick the right stack and configuration
AI stack can be expensive, but manageable if you clearly define the use case you're solving.
For example:
- We’ve repeatedly solved problems using the lowest-priced model (e.g., gpt-4o-mini).
- It did require intense prompt engineering—but that’s far cheaper.
- Same with RAG—though that deserves a separate post.
Use AI for all of your business
We use AI not just for our product, but across the business:
- Development: we use Cursor heavily.
- Communication: all of us are power users of ChatGPT, Claude, and Gemini.
- Documentation & customer emails: AI helps us draft quickly and accurately.
So to all entrepreneurs sitting on the sidelines:
This is the best time to build a company that can challenge incumbents and outdated workflows—without raising tons of money—and have fun while doing it.
