This isn’t your average "volume-first" sales role. We need a GTM Engineer who treats pipeline generation as a technical craft.

• The Vibe: You own the pipeline from initial sourcing to the final close.

• The Stack: Signal-based prospecting, automated workflows, and multi-channel outreach.

• The Goal: Moving past generic templates to build thoughtful, scalable, and high-conversion outbound processes.

2. Senior Product Engineer

We’re looking for a builder who loves the "new way" of coding but respects the "right way" of engineering.

• The Vibe: You’re an AI-native developer (Cursor/Claude Code are your best friends) who can ship features in hours, not weeks.

• The Edge: Strong product intuition. You don't just "take tickets"—you think like the user and own delivery end-to-end.

• The Rigor: Speed is a feature, but reliability is a requirement. You build systems that are robust, scalable, and production-ready.

Please DM to share your resume or send it to him directly here : narayan@metaflow.life

Open Roles :

1. Data Engineer (Capital Markets)

Help a leading financial technology firm manage complex balance sheets through innovative data solutions. This is a hands-on role where you will design and deploy end-to-end data delivery systems.

• Experience: 7–8 years in data management and business intelligence.

• Tech Stack: Proficiency in Python, Snowflake, SQL Server, and PowerBI.

• The Mission: Architect and implement data warehouses, lakes, and marts while leveraging cloud services like AWS or Azure.

• Work Timing: Shift follows UK hours (13:30 – 23:00 IST).

2. Senior Software Engineer (Data Science/ML)

Join the core technology team of a Retail Tech firm to shape the foundation of products that directly impact how modern retailers operate.

• Experience: 4–6 years developing and deploying production-ready ML solutions.

• Tech Stack: Strong proficiency in Python and core libraries like Pandas and NumPy.

• The Mission: Design and build ML pipelines—from data preprocessing and feature engineering to deployment and monitoring.

• Innovation: Work on optimisation and Generative AI to bring practical innovation to the product roadmap.
  
  Why Join?
  
  These positions offer the chance to own the "full stack" of data, from sourcing to final analysis, while collaborating with global teams of talented professionals.

Sharing the full JD in attachments along this post.

With India’s DPDP Act imposing penalties up to ₹250 Crore, "moving fast and breaking things" could cost you everything. If you are building AI, SaaS, or data-driven startups, understanding these regulations is no longer optional—it's survival.

🧠 The Topic

The ₹250 Crore Question: What DPDP Actually Means for Your AI Product We’re breaking down the complex legal landscape into actionable insights for founders and builders to ensure your innovation remains compliant and scalable.

🎙️ The Experts

• Speaker: Sarthak Dash Bhattamishra, a veteran Technology Lawyer and Policy Commentator with deep expertise in AI governance and CIPP/E certification.

• Host: Btrained (an initiative by Baigen Technolabs), dedicated to providing practical, operations-first learning for the next generation of tech leaders.

• AI Brains is excited to partner with Baigen Labs to host this event.
  

  
  📅 Details
  

  • Date: March 14, 2026

  • Time: 11:00 AM

  • Platform: Online Webinar

  • Register here : https://forms.gle/zTnfwedgRt61d4eE7

The Core Takeaway:

Context is King : Anjan broke down the true power of Claude Co-work, the evolution of Claude Code. The key differentiator is its ability to process entire folders and documents across different folders, overcoming the file-size and context limitations of earlier tools.He showcased how this enables real automation for non-technical users through a clean desktop app interface, highlighting features like the "Progres"s area (which breaks down complex tasks) and custom Skills that can anchor content creation to a user’s unique thinking style.

Practical Use Cases Demonstrated:

• File Organization: Cleaning up a "messy downloads" folder instantly with descriptive file names.

• Meeting Workflow: Summarising complex meeting notes, extracting action items, and drafting targeted follow-up emails for different stakeholders, all in one go.

• Content Creation: Using a pre-built "thought leadership skill" to draft a social media post, referencing multiple documents.

• Automation: Automating daily team summaries (/slash today reports) and expense processing by connecting to Linear/Gmail.

If you’re looking to move beyond simple chat prompts and integrate AI into your daily operational workflow, this recording is a must-watch!

🔗 Watch the full recording here:

Discover the exact prompts and workflows Anjan used to bridge the gap between complex AI tools and everyday productivity.

This is for you if:

• You are a full stack engineer who has shipped products at scale.

• You want to build voice AI applications from scratch.

• You thrive on crossroads of serious coding and vibe coding.

• You have a "hacker" mindset but "architect" discipline.

• You see the Voice AI explosion and want to be the one holding the fuse.

• You love startups and a person with high agency.

Brownie Points : Those who have experience building voice AI apps previously will be given top preference.

The Reality: We aren't looking for a "manager" who sits in meetings; we’re looking for a builder who wants to own the technical roadmap and eventually lead the entire org.

Please use the link below to submit your resume

https://forms.gle/XgD1SSzikh6vvHHNA

• Co-creator of the first popular web browser (Mosaic) in the early 1990s, which helped bring the internet to the public.

• Co-founder of Netscape, whose browser played a major role in the early web boom.

• Co-founder of Andreessen Horowitz (a16z), one of the world’s most influential venture capital firms.

• Co-creator of the first popular web browser (Mosaic) in the early 1990s, which helped bring the internet to the public.

The manifesto argues that progress slows not because of technical limits, but because of fear-driven decision-making. As builders, this is a reminder that:

• Our responsibility is to solve real problems, not avoid them

• Speed and experimentation matter, but so does accountability

• AI and automation should be seen as leverage for human capability, not replacement.

• Poverty, disease, and low productivity were the default state of humanity

• Technology not luck helped overcome them

• Slowing innovation may unintentionally slow social progress

The manifesto also introduces an important distinction:

• Optimism ≠ ignoring risks

• Optimism means believing problems are solvable through effort and learning.
  
  
  Topic's for Open Discussion,
  

• Are we solving problems that truly matter, or just safe ones?

• Are we letting public fear shape our product decisions?

• Are we optimistic enough to bet on progress even when outcomes aren’t guaranteed?
  

Some questions that will help to think in the right direction.

1. Strategic Foundation

• Have I prioritized the right use case?
  A good use case for an agent involves nuanced judgment, unstructured data, or rules that have become too complex for traditional automation.

• Have I established a performance baseline?
  Prototyping with the most capable model available sets a benchmark before optimizing for cost or latency.

• Have I justified the orchestration pattern?
  Having clarity on when you need single-agent system. A multi agent system is required if the prompt is overwhelming and the logic is hard to contain in a single go.

2. Core Components & Logic

• Have I standardised my tools?

  Create well-documented, reusable tool definitions that allow for flexible, many-to-many relationships across different agents.

• Have I grounded my instructions in reality?
  Always use existing company SOPs, support scripts, or policy documents to create the agent's routines.

• Have I eliminated ambiguity in steps? I
  Ensure every instruction in the agent’s routine corresponds to a specific, unambiguous action or output.

• Have I mapped out edge cases?
  Build conditional branches to tell the agent exactly how to handle missing information or unexpected user inputs.

3. Safety & Guardrails

• Have I implemented a layered defense?
  I am using a combination of LLM-based classifiers, rules-based regex, and moderation APIs to vet all inputs and outputs.

• Have I rated my tool risks?
  I have assigned risk levels (low, medium, high) to every tool and mandated extra checks or human approval for high-risk actions.

• Have I optimized for latency?
  Enable "optimistic execution" so that guardrails run concurrently with the agent's response generation. Latency for different use cases is regularly calcuated.

4. Human-in-the-Loop

• Have I set failure thresholds?
  I have defined the exact number of retries or failures allowed before the agent must automatically stop and hand off to a human.

• Have I protected high-stakes actions?
  I have ensured that any irreversible or sensitive task requires a human to "green-light" the execution.

The agent's success largely depends on the maturity of the tool calling setup made available to it also, this is the first step of production readiness. Specially if your Agent is expected to do dynamic actions. Where most agents today fail :

• Dynamic Data: RAG is often "static" (retrieving fixed documents). It cannot perform real-time calculations or fetch a customer's specific balance from HubSpot without a tool.

• System Action: Prompts can say they want to cancel a ticket, but they can only do it if connected to a tool API like Freshdesk.

• Precision Extraction: RAG struggles with complex tables or highly structured pricing rules; these are better handled by structured code or "Custom Actions"

Oftentimes, premature setup, stale data and lack of onboarding makes most internal tools unfit for your Agentic automation stack.

Striving for Organisational clarity

The guardians of these tools and the teams using them usually live in different towers. Effective implementation calls for a fluid work relationship between these teams. Organisational readiness for AI agents can be done in the following ways.

1. The "Tool Audit" Workshop

• Goal: Identify what the agent actually needs to do.

• Action: Have subject matter experts (SMEs) list the 10 most common manual tasks. For each task, map it to a specific system (HubSpot, Gmail, etc.) and identify if an API exists or if "computer-use" models are required.

2. Implement "LLM-Friendly" Documentation

• Problem: Traditional API docs are written for human developers, not LLMs.

• Solution: Create a simplified "llms.txt" or a text-first digest of your internal APIs. Strip away jargon and provide clear "if-this-then-that" examples for the agent to follow.

3. Establish an AI Center of Excellence (CoE)

• Role: A cross-functional team (Engineering, Legal, and Business) that sets the "Risk Appetite" for the organization.

• Outcome: This group defines which tools are "auto-approve" and which require a human "emergency exit".

4. Systematic Evaluation (Evals)

• Method: Move away from "vibe-based" testing to systematic evals.

• Action: Create a "test set" of user prompts and record whether the agent chose the correct tool. Use an "LLM-as-a-judge" to grade the reasoning behind the tool choice.

What are your learnings/ ideas and experience with building AI agents.
Do you have any specific tips for the community that can benefit anyone building AI agents?
What are some of the top challenges you face in building production grade AI agents?

Share your thoughts in comments.

A self inventory checklist would look like this :

• Logic: Have I given the model control over the workflow, or is it just following a fixed script?

• Knowledge: Have I turned my messy docs into clear, step-by-step instructions for the model?

• Tools: Are my tool definitions clear enough that the model knows exactly when to use them?

• Scale: Have I pushed a single agent to its limit before trying to build a complex multi-agent system?

• Safety: Do I have automated checks running to catch bad inputs or dangerous outputs?

• Escalation: Does the agent know exactly when to stop and ask a human for help?

From my working observation of Agents, Prompting and RAG (Retrieval-Augmented Generation) together handles intelligence and context, while tool calling handles the real action. Check our previous blog to learn more about Agents. Most people might think Prompting solves most of the challenges, but oftentimes it's not. If you are building an Enterprise grade AI Agent - there's more. Tool calling needs to be standardised and mature for the agent to perform well.

What's your experience with building AI Agents, do share your personal checklist on what works.