9. AI Brains Weekly (Apr 1- 15) .

Our new website went live on April 14. Designed using Google Stitch and built in Framer. The visualisation aspects of Stitch was promising but we ended up using Framer for repeatability, SEO, control and other nuances. Once you design the website, you are able to export the design files in different formats or extract a HTML file too.

Note : Mobile version of the website still in the works.

Based on the community poll, we've launched 5 focus groups in Whatsapp: Agentic AI, AI Assisted GTM, AI Agency, AI Research Focus, and a group for AI Engineers, FDEs, and CTOs. The objective is to map all the important hiring, fundraising, teaching, play books, insights and offers in a single place for everyone to benefit. Someone needs to identify the top minds operating in AI in our space and we are taking up this task for the wider community to benefit. We are also working on other upcoming topics shown,

If you would like to be part of these group of highly pointed discussions, please take a moment to fill this form here.

We’ll find the right group to post your problems and challenge.

Hiring Alerts :

Anjan Panneer Selvam is looking for an early-career PM, comfortable with advanced AI tooling, ideally familiar with Indian law terminology, and ready to move fast. DM Anjan directly if you know someone.

4.1. Multi-LLM workflows are becoming standard

Santhosh, from Humcode shared a two-model setup getting traction: use Claude Code for architecture and planning, produce a detailed markdown spec with tests for each phase, then hand that spec to Codex for execution. The rationale is clean. Claude designs more holistically. Codex stays within a tight plan reliably. The broader question this opened was whether we are still thinking about AI tools individually when we should be thinking about them as a coordinated stack.

4.2. Vibe coding vs production engineering

This was a 🔥 topic, that ran for 2 full days. Most of them agreed that vibe coding is powerful for getting to a prototype fast, but DB optimisation, security, caching, load balancing, observability, and system design are still fundamentally human engineering decisions. Sandeep, from Baigen Technolabs, flagged something worrying from recruiting candidates who built apps using AI but cannot answer basic questions about the libraries they used or how their systems actually work.

Vineeth Vijayaraghavan, made the sharpest distinction of the week. Vibe coding, copilot tools, and agentic AI coding are three completely different categories. Most of the noise in this conversation comes from people who have not separated them.

4.3. The AI provider dependency problem. Moving towards Self hosted Infra.

Last night, Suraj laid out a question many of us carry quietly. What happens when your entire business, context, memory, client projects, dev workflow, lives inside one AI provider? His team's response has been a hybrid setup: Sonnet 4.6 as orchestrator, Qwen 3.5 as executor inside Hermes agents. Anjan's team took a different approach, adopting Karpathy's documentation framework to build context that is portable across providers and owned by no one. The community consensus: this is not a future problem. It is a present one. A community online meetup on self-hosted infrastructure and provider portability is being planned.

4.4. Claude degradation

Multiple members flagged this independently across both weeks. Claude Rate limits are being hit faster. Quality is dipping on tasks that used to be reliable. A Claude Code memory handling bug in early April also caused a spike in token consumption. Interest in self-hosted setups and open-source alternatives is noticeably higher. One of our members observed about Claude issues.

Sridevi Prabhu, launched ProMarshal, the base version of her AI-powered project management product, months in the making, now live. She is actively seeking feedback and bug reports.

Suraj built a Twitter growth agent using Hermes that reads his timeline, identifies relevant conversations, drafts replies in his voice, and posts after his review. Targeting 20 daily replies engaging top AI voices on X. He is building the same for LinkedIn next.

Nelson M Sathya is building an AI-powered, assessment-gated apprenticeship marketplace focused on Tier 2 and Tier 3 talent. They are early stage, equity-based, looking for a co-founder who can build and move fast. DM Nelson if this is a problem that resonates well with you.

Narayan, CEO of MetaFlow’s take on AI sales agents like Swan, 11x, and similar agent companies are not selling into SaaS budgets. They are framing it as a headcount and payroll conversation. This changes how you sell and how you think about pricing these tools as opposed to SaaS pricing.

This is a section on some interesting posts, links and shares from the community

Join the AI Brains community and be part of these conversations. See you on next week’s catch up.

Note : None of the tools, tweets, people mentioned are being promoted here. This is a simple log of what’s happening in the community for the broader folks who follow our updates regularly. Thanks