DevRelCon NYC 2026 Lineup: Agents, Adoption, and Developer Trust

🚀 Building the Future of B2B Client Onboarding: If you’ve ever worked in a B2B agency (marketing, accounting, legal), you know the absolute pain of onboarding new clients. It usually involves endless back-and-forth emails, missing documents, fragmented spreadsheets, and answering the exact same procedural questions twenty times over. The result? The average onboarding time stretches out to ~21 days, draining internal team bandwidth and stalling time-to-value for the client. I’m currently building Intelligent Client Onboarding Agent to solve exactly this. 💡 The Solution A multi-tenant SaaS platform where agencies can upload and train an autonomous AI agent directly on their Standard Operating Procedures (SOPs). The agent acts as a dedicated 24/7 onboarding concierge for new clients: Collecting and validating required document uploads. Answering complex procedural questions instantly using RAG (Retrieval-Augmented Generation). Compiling a final, comprehensive brief for human approval once the client is ready. By moving the heavy lifting to an AI agent, we can crash average onboarding times from 21 days to under 5 days, reducing manual staff effort by 80%. 🛠️ The Tech Stack To handle complex AI orchestration, async processing, and multi-tenant isolation, I chose a robust enterprise-ready stack: Frontend: React.js + Tailwind CSS for a seamless, responsive UI. Backend API: Django + Django REST Framework (perfect for robust data modeling and security). Async Workers: Celery + Redis to manage background AI reasoning and processing. Vector Database: ChromaDB to store and query chunked SOP documentation. AI Orchestration: LangChain for managing autonomous agent tool execution and LLM contexts. Database & Infrastructure: PostgreSQL for transactional data, fully containerized using Docker. 🏁 Phase 0 Complete! I have officially laid down the core architecture foundations: Implemented secure Authentication & Tenant Scaffolding to ensure absolute data isolation between agencies. Containerized the entire local development environment using Docker & Docker Compose. Integrated Groq for high-speed LLM inference to power the initial agent interaction loops. Next up: Deepening the RAG pipeline and building out the dynamic document collection engine! Check out the progress or explore the project architecture on GitHub: Intelligent-Client-Onboarding-Agent #SaaS #AI #Django #ReactJS #LangChain #Solopreneur #WebDevelopment #AIagents #BuildInPublic

I've shipped 50+ SaaS products, scaled a 300-person team, and sent millions of cold emails. The lesson underneath all of it wasn't technical. It was this: most of what gets taught about AI and B2B growth is written by people who've never had to ship the thing. I have. Repeatedly. In production: real clients, real deliverability problems, real teams to run. I'm Daksh. Founder of Sortwind (custom SaaS, CRMs, and AI workflows) and co-founder of MX Validator (email deliverability and validation for B2B outreach). Before that, I ran a 300-person cross-functional team across HR, engineering, and marketing. I write Node.js, Python, and MERN, and I've spent years inside the unglamorous plumbing of cold email list hygiene, sequencing, and the deliverability failure modes nobody posts about. This page is where I document the real version of building: - AI workflows that survive contact with an actual client - The SaaS blueprint: architecture, shipping, launches, the parts that break - Cold email and B2B growth from the deliverability layer up - The operator's MBA: what running 300 people teaches you that no course does No fluff, no "revolutionize your workflow." Just the operator's view, from someone still in the arena. Follow along if you'd rather learn from what shipped than from what sounds good on a slide. To start: What's the biggest gap you see between people who teach building and people who actually build?

↓ 10 open-source repos so good they shouldn't be free: Most people pay hundreds a month for tools that already exist, open source, often better, and yours to keep. Here's the stack: A real browser for your agents, so they actually click and fill forms. An open coding operator you self-host instead of renting. A self-hosted ChatGPT-style UI for your whole team. Your own Vercel, Heroku and Netlify in one. A backend that quietly replaces five separate bills. PDF superpowers with no Adobe tax. Plus tools that turn the web into clean data your agents can use. Between them, they kill hundreds a month in subscriptions. Here is the catch nobody mentions: ten repos across ten dashboards is its own kind of mess. So I wired all of them into Ultron, one workspace, one system, agents that actually use them together instead of ten things you forget you're running. Own your stack. Stop renting it. Comment UNLOCK and I'll send you the install and setup guide for all 10. #fyp #ai #automation #startup #growth

Just read a comparison of OpenCode and OpenClaw and realised most people are asking the wrong question about AI agents. They're not competing tools. They're solving completely different problems, and conflating them is how you end up spending money on the wrong thing. OpenCode is a coding agent. You point it at a repo, give it a specific task like "fix this failing test" or "refactor the auth middleware, " and it reads files, makes a plan, edits code, runs tests. It lives in your project. OpenClaw is an always, on assistant gateway. It connects to Slack, Telegram, Discord, WhatsApp, whatever. It handles scheduled tasks, browser automation, persistent workflows. It lives between you and your tools, waiting for messages. I've watched clients burn money by buying the fancy always, on setup when they just needed a focused coding agent for their CI/CD pipeline. And I've seen others try to use a repo, focused tool for things it was never designed to do. The real cost trap with both of these isn't the tool itself. It's context bloat. MCP tools pile up. Channel access creeps. Workflows get loose. Suddenly your model calls are 10x what you budgeted because nobody actually scoped the permissions. Same lesson I learned the hard way with Stripe integrations back in the day. You don't pay for the feature. You pay for the feature you didn't think through. Which one actually fits what you're trying to automate? That's the only question that matters. https://lnkd.in/d8Ax5RvN

These are dangerous times we live in. But, weren't they always? Even before AI, big brands were dictating how to play the software development game. "Don't use that, use this." "There's a sucker born every minute, and you're right on time." These two quotes pop into my mind each time there's a new fab that everybody talks about. We had a new web development framework pop out every few moments back in the day, promising to revolutionize the field in ways you couldn't even imagine. And here comes Salesforce, joining the game, perhaps a bit late. And there come the blog posters, "software developers," and "Salesforce evangelists". What does their new shiny AI tool do? - Nothing new. What do bloggers write about? - How is this new shiny version better than v1 you tried and did not like, praising it, saying you should give it another try. What do they promise? - Revolution What do they deliver? - Reinventing the wheel. Each feature is praised under a pseudonym. You ask what that pseudonym stands for. The very start of a descriptive paragraph provides a comparison with an existing Claude or Cursor feature. What do they expect people to do? - Well, it's a Salesforce, and you don't trust anybody else but us, right? Claude and Cursor are 3rd parties; are they the enemy? Join us - resistance is futile.

Over the last month I built Pro Engineer, an AI support-engineering assistant inside our xCloud Slack workflow. The problem was familiar to anyone in support-heavy SaaS. Developers spend a lot of time not coding. They spend it context-switching: 🔍 Finding which server or site an issue is on 🛠️ Tracing the right code path 📋 Reading logs safely 🐞 Deciding if something is a support action, expected behavior, or a real bug That work is draining. It pulls focus and slows everyone down. Pro Engineer handles that first investigation layer right in Slack, before a developer gets pulled in. This is not about replacing developers. They still step in, but later, with better context and a cleaner request. Support gets faster direction. Devs face fewer vague interruptions. Customers get a more consistent path to resolution. What makes me proud is how grounded it is. This is no generic chatbot bolted onto Slack. It connects to the real xCloud setup: ✅ Product-specific support contexts ✅ Safe database views ✅ GitHub issue and PR context ✅ Escalation rules ✅ Support-friendly steps It changed the flow from "support asks a dev to investigate" to "AI prepares the investigation, support acts when safe, and devs step in only for a real product or code decision." I led the build of Pro Engineer end to end, from the Slack integration to the safe database layer and escalation logic, as part of the wider AI work the team has been pushing at xCloud. I improve it every week, and the early signal is strong. The key numbers are in the graphic above. 📊 Proud of what we build at xCloud. 🚀 💬 Are you working on workflows that use agents or tools for developers? If so, comment below. #AgenticEngineering #WorkflowAutomation #AI #DeveloperProductivity #CustomerSupport #SaaS #EngineeringLeadership #xCloud #AIWorkflows

Most teams we talk to are stuck in the same place: they need modern web, mobile, AI, or automation shipped faster than they can hire without a throwaway build from a template shop. EatCodeSleep is an engineer-led agency under AI · Software · Automation. We design and build scalable platforms web apps, mobile, AI integrations, agents, and automation workflows — then stay on for growth support. We also work on SaaS products (Stax Fun, Qreates, Croptalk, and Clipflow), so when we talk AI product delivery, it is practice, not a slide deck. What that usually means for a team: → Validate and ship an MVP → Modernize an existing stack → Add LLM / RAG / agents → Wire automation that cuts manual ops We run discovery → architecture → build → deploy → support — as a delivery partner or as an extension of your eng team. Accepting projects for Q3 2026. If that sounds useful: 30-minute discovery. Link in comments / bio. 🌐 https://lnkd.in/dXeSw3_z #AI #SoftwareAgency #Automation #SaaS #BuildInPublic

🚀 Excited to share one of the products I recently built — QualifyChat, an AI-powered Sales Agent platform designed to help service businesses turn website visitors into qualified leads 24/7. Instead of losing potential customers after business hours, QualifyChat engages visitors instantly, answers questions, qualifies leads, captures contact information, and helps businesses convert more opportunities automatically. 💡 What I worked on: ✅ Product Architecture & System Design ✅ React Frontend Development ✅ Node.js Backend Development ✅ AI & LLM Integrations ✅ Real-time Chat Functionality ✅ Multi-Tenant SaaS Architecture ✅ Database Design & API Development ✅ Dockerized Deployment Infrastructure Key Features: 🔹 AI-Powered Conversations 🔹 Lead Qualification & Capture 🔹 Real-Time Dashboard 🔹 Website Widget Integration 🔹 Multi-Tenant SaaS Platform 🔹 Automated Customer Engagement Building products from idea to production is always an exciting challenge, and QualifyChat was a great opportunity to combine AI, SaaS architecture, scalable backend systems, and user-focused design into a single platform. Always looking forward to building innovative products that solve real business problems. Here is the link 🔗https://lnkd.in/dHR_8Tue #FullStackDeveloper #SoftwareEngineer #ReactJS #NodeJS #TypeScript #SaaS #ArtificialIntelligence #OpenAI #LLM #WebDevelopment #ProductDevelopment #Docker #PostgreSQL #Startup #TechInnovation #LeadGeneration #Automation #AIProducts #QualifyChat

Here's a Slack+Zapier+Claude set up where the team can ask "analyze my last calls with MEDDICC" in a Slack channel. The agent pulls recent call summaries from your conversational intelligence tool, reasons over them with AI, and replies in the thread. Each follow up builds on the last without losing context Where your team chats is a natural place to call up on an agent The original build uses five visual Zapier steps and a Vercel function as two separate services. Zapier SDK collapses that Here's how it works 𝗪𝗵𝗮𝘁 𝗭𝗮𝗽𝗶𝗲𝗿 𝗦𝗗𝗞 𝗶𝘀 Governed access to 9,000+ app integrations, callable from a script a coding agent writes. All of the hard stuff is handled by Zapier: auth, token refresh, retries, and error handling all built in. You're calling app actions from code, not from the visual Zap builder 𝗧𝗿𝗶𝗴𝗴𝗲𝗿 𝗹𝗮𝘆𝗲𝗿 Slack message fires the agent (easy!) 𝗠𝗲𝗺𝗼𝗿𝘆 𝗹𝗮𝘆𝗲𝗿 You could still leverage Zapier Storage for statefulness. Or you can also leverage the SDK, which opens the door to Zapier Tables, which is more structured, queryable, and easier to debug when something breaks 𝗖𝗼𝗺𝗽𝘂𝘁𝗲 𝗹𝗮𝘆𝗲𝗿 My original setup involved Zapier and Vercel. Zapier SDK removes that constraint because it collapses it all into one script (the conditional call summary fetch, the system prompt build, the Claude call) Vercel can still be the deployment target. It's no longer the only one 𝗗𝗮𝘁𝗮 𝗹𝗮𝘆𝗲𝗿 Your conversational intelligence tool connects through the SDK. Auth is handled automatically. The decision to pull summaries over full transcripts is an architectural call that carries over regardless of how the orchestration is wired 𝗖𝗮𝗰𝗵𝗶𝗻𝗴 𝗹𝗮𝘆𝗲𝗿 No change. The CLAUDE.md caching logic lives in the Claude API call. The orchestration layer around it is irrelevant 𝗥𝗲𝘀𝗽𝗼𝗻𝘀𝗲 𝗹𝗮𝘆𝗲𝗿 Slack reply still posts in the thread using thread_ts. SDK handles Slack auth. No manual token management The SDK collapses everything into one flow My original build was an elegant five step Zap. Now it's just one script with SDK. The Vercel function becomes optional. The conditional logic that required a separate compute layer is now an if statement with error handling built in For the no code builders: Zapiers SDK is the way to go Go forth and operate 👋

→ 𝐌𝐨𝐬𝐭 𝐃𝐞𝐯𝐓𝐨𝐨𝐥𝐬 𝐃𝐨𝐧’𝐭 𝐅𝐚𝐢𝐥 𝐨𝐧 𝐏𝐫𝐨𝐝𝐮𝐜𝐭. 𝐓𝐡𝐞𝐲 𝐅𝐚𝐢𝐥 𝐨𝐧 𝐃𝐢𝐬𝐭𝐫𝐢𝐛𝐮𝐭𝐢𝐨𝐧 𝐒𝐭𝐫𝐚𝐭𝐞𝐠𝐲 A strong tool without a growth system stays invisible. In developer ecosystems, discovery is the real bottleneck. 𝐇𝐞𝐫𝐞’𝐬 𝐡𝐨𝐰 𝐫𝐚𝐧𝐤𝐢𝐧𝐠 𝐚𝐧𝐝 𝐚𝐝𝐨𝐩𝐭𝐢𝐨𝐧 𝐚𝐜𝐭𝐮𝐚𝐥𝐥𝐲 𝐜𝐨𝐦𝐩𝐨𝐮𝐧𝐝: • 𝐃𝐞𝐯𝐞𝐥𝐨𝐩𝐞𝐫-𝐅𝐨𝐜𝐮𝐬𝐞𝐝 𝐒𝐄𝐎 (𝐃𝐢𝐬𝐜𝐨𝐯𝐞𝐫𝐲 𝐋𝐚𝐲𝐞𝐫) ✓ Target high-intent search behavior across Google and GitHub ✓ Build comparison pages that capture evaluation-stage traffic ✓ Align docs + repos + content for unified discoverability • 𝐇𝐢𝐠𝐡-𝐕𝐚𝐥𝐮𝐞 𝐓𝐞𝐜𝐡𝐧𝐢𝐜𝐚𝐥 𝐂𝐨𝐧𝐭𝐞𝐧𝐭 ✓ Use-case driven narratives instead of generic blogs ✓ Tutorials and comparison content that mirror real engineering decisions ✓ Documentation treated as product surface, not support material • 𝐏𝐫𝐨𝐝𝐮𝐜𝐭 𝐃𝐨𝐜𝐮𝐦𝐞𝐧𝐭𝐚𝐭𝐢𝐨𝐧 𝐎𝐩𝐭𝐢𝐦𝐢𝐳𝐚𝐭𝐢𝐨𝐧 ✓ Reduce time-to-first-success aggressively ✓ Add real-world implementation patterns and SDK clarity ✓ Treat onboarding as conversion, not information • 𝐆𝐢𝐭𝐇𝐮𝐛 𝐆𝐫𝐨𝐰𝐭𝐡 𝐒𝐭𝐫𝐚𝐭𝐞𝐠𝐲 ✓ README as positioning layer, not just instructions ✓ Demos, templates, and runnable examples as trust accelerators ✓ Contribution loops designed for organic traction • 𝐂𝐨𝐦𝐦𝐮𝐧𝐢𝐭𝐲-𝐋𝐞𝐝 𝐆𝐫𝐨𝐰𝐭𝐡 𝐄𝐧𝐠𝐢𝐧𝐞 ✓ Build developer spaces around problems, not products ✓ Structured engagement loops via feedback and discussions ✓ Turn contributors into distribution channels • 𝐃𝐞𝐯𝐞𝐥𝐨𝐩𝐞𝐫 𝐄𝐱𝐩𝐞𝐫𝐢𝐞𝐧𝐜𝐞 𝐎𝐩𝐭𝐢𝐦𝐢𝐳𝐚𝐭𝐢𝐨𝐧 ✓ APIs designed for predictability and low cognitive load ✓ SDKs, CLI tools, and integrations that reduce friction density ✓ Every reduction in complexity increases adoption probability • 𝐏𝐫𝐨𝐝𝐮𝐜𝐭-𝐋𝐞𝐝 𝐆𝐫𝐨𝐰𝐭𝐡 𝐒𝐲𝐬𝐭𝐞𝐦𝐬 ✓ Self-serve onboarding replaces dependency on sales cycles ✓ Freemium structures aligned with expansion triggers ✓ In-product nudges aligned with usage milestones • 𝐃𝐢𝐬𝐭𝐫𝐢𝐛𝐮𝐭𝐢𝐨𝐧 𝐋𝐚𝐲𝐞𝐫 𝐄𝐱𝐩𝐚𝐧𝐬𝐢𝐨𝐧 ✓ Launches and listings as early visibility multipliers ✓ Developer communities as sustained acquisition channels ✓ Strategic presence across curated ecosystems and forums • 𝐀𝐧𝐚𝐥𝐲𝐭𝐢𝐜𝐬 & 𝐅𝐮𝐧𝐧𝐞𝐥 𝐈𝐧𝐭𝐞𝐥𝐥𝐢𝐠𝐞𝐧𝐜𝐞 ✓ Track drop-offs across developer journeys, not just signups ✓ Optimize activation and retention signals continuously ✓ Use behavioral data to refine onboarding loops • 𝐅𝐥𝐲𝐰𝐡𝐞𝐞𝐥 𝐀𝐜𝐜𝐞𝐥𝐞𝐫𝐚𝐭𝐢𝐨𝐧 ✓ Users evolve into contributors and advocates ✓ Real-world use cases become acquisition assets ✓ Open-source participation strengthens ecosystem lock-in → In DevTools, product quality creates retention. But distribution architecture determines survival. P.S. Curious to understand how teams are balancing product depth with distribution engineering in today’s developer-first landscape. Follow Shantanu Das ↗️ for more insights