Tech Stack Management

If you’re overseeing an Agentic AI roadmap, these ten principles can save cost, carbon, and complexity. In the race to deploy autonomous agents, many organizations are quietly accumulating Agentic Debt — systems that are over-orchestrated, expensive to run, and increasingly hard to govern. Engineering excellence in the AI era isn’t about how much autonomy an agent has. It’s about how much efficiency, restraint, and intent are baked into the architecture. Here are the 10 Lean Agentic AI Principles for building production-ready, sustainable systems: 1. Managed Context – Large context is a liability when unmanaged. More memory ≠ more intelligence. 2. Right-Sized Models – Not every prompt deserves a 70B response. Use the smallest brain that gets the job done. 3. Streamlined Orchestration – Agent orchestration is not a playground. Every extra agent is a cost, a delay, and an emission. 4. Think Before Compute – Reflections aren’t free. Validate the need before asking an agent to “think.” 5. Targeted Retrieval – RAG isn’t always right. Retrieve only when it’s truly needed. 6. Account for Hidden Emissions – Emissions don’t show up in logs, but the planet still pays for them. 7. Reuse as Reasoning – Don’t re-run. Re-think. Reuse is the new reasoning. 8. Judicious Tool Use – More tools, more problems. Every tool adds latency and risk. 9. Judgmental Memory – Memory isn’t a journal. Storing everything is hoarding, not intelligence. 10. Governance Over Autonomy – Agentic systems need governance. Left unchecked, autonomy becomes chaos. A lean mindset doesn’t just reduce overhead. It increases predictability, performance, and trust across the entire agentic stack. These ideas are now open-sourced as the Lean Agentic AI Playbook: https://lnkd.in/dp8KZVku. For deep dive , refer to my book - https://leanagenticai.com/ #AgenticAI #LeanAgenticAI #SustainableAI #SoftwareArchitecture #AIStrategy #ResponsibleAI

You didn't build a 'modern data warehouse.' You built the world's most expensive junkyard. I recently audited a client's infrastructure with 120+ tables loading every night. -Not one person could explain what half of them were for. -Dashboards contradicted each other. -Analysts were burning hours tracing dependencies instead of building anything useful. -If few engineers left everything would crumble. On the top the cloud bill and data volume is growing. Here’s the problem no one wants to admit Your ELT isn’t broken. It’s bloated. We’ve confused “accessibility” with “intent” Tools made it easy to load everything so we did. And over time, your data warehouse turned into a junkyard. The Hidden Cost of Loading Everything Every table you load has a cost: Storage & compute. (That Snowflake bill didn’t double by accident.) Engineering time. (Maintaining pipelines no one uses.) Trust. (Conflicting numbers, different definitions, zero confidence.) Everything you add to tech stack can and most likely will become a liability faster then the asset. The result? You’re sitting on a goldmine of data but its usless Creating 1. Start with Decisions, Not Sources Every dataset should answer a business question. If you can’t tie it to a KPI, a metric, or a decision it doesn’t deserve a pipeline. Rule: “If we can’t name who uses it or what decision it drives, delete it.” 2. Audit Everything You Load Run a two-hour audit of your pipelines. Tag each as: Must-Have: Directly drives a core KPI or compliance requirement. Nice-to-Have: Useful but infrequently used. Archive: Unused in 90+ days. When we did this for a Healthech and finance client, 40% of their pipelines had no active usage. They were maitaining duplicate pipelines, cost a fortune but nobody haven't even looked at 3. Connect Every Pipeline to the P&L Nobody cares many pipelines you’ve built they care how it impacts margin. Every data initiative should tie to one of three levers: Cost reduction (cloud spend, engineering time) Revenue enablement (forecast accuracy, churn prevention) Risk reduction (audit accuracy, compliance) If it doesn’t hit one of these? It’s noise 4. Assign Ownership The most expensive part of pipeline and ETLs isn’t compute. It’s ambiguity. No one owns the data. No one knows who built it. No one knows what engineer responsible for it. Assign a data steward per domain responsible for purpose, lineage, and consumers. Accountability drives cleanup faster than any governance tool. 5. Enforce an “ROI Gate” Before Every New Ingestion Before any new source is added, ask: “What decision does this support, and what’s the expected ROI?” You’ll kill 50% of waste before it even starts. Every unnecessary dataset or tools you load burns margin, trust, time and complexity. Most likely a liability. Build lean, trusted, scalable and AI-ready data architecture. Stop loading everything. Start loading with intent.

Building an AI-native company with 2 people. Here's the exact stack running it. Four weeks ago I started sharing the systems inside our company. A lot of you asked: "What's the actual stack?" Here it is. The brain: Claude Code. Every system I've described was built in coding sessions with AI. Not vibe-coded. Directed. I write detailed specs with micro-tasks, then execute them methodically. The database: Supabase. Postgres with row-level security. Clients, deals, contacts, actions, notes, activity logs, proposal outcomes, engagement health. All in one project. The frontend: Next.js with React, a component library, and Tailwind. Deployed on Vercel. The glue: Not Zapier. Not Make. Python scripts and TypeScript sync scripts that run on cron. The scripts are simple, 50-100 lines each. The power is that they all share the same database. The agents: 7 role-based AI agents that run on schedule. Inbox manager, pipeline checker, daily summary. Communication: Slack for internal updates, Telegram for personal tracking, Gmail drafts via IMAP. Client delivery: Airtable for content tracking, Notion for client-facing knowledge bases, Google Workspace via CLI. Total monthly cost: basically zero. Free database tier. Free hosting tier. One AI subscription. The takeaway: you don't need a team to build real infrastructure anymore. You need clarity on what you want, the patience to build it piece by piece, and an AI that can code. What's your stack for running lean?

Your rev ops team is drowning in tool implementation instead of driving revenue I watched a company spend 8 months implementing their "perfect" sales tech stack. Salesforce. Outreach. Gong. ZoomInfo. Drift. Calendly. LeanData. PandaDoc (some of these they had already and others they bought in the year) They finally finished the rollout and celebrated with an all-hands meeting about their "modern revenue engine." Six months later, their sales productivity was down 13%. The problem wasn't the tools. It was the philosophy. They optimized for features instead of outcomes. While they were building the perfect tech stack, their competitors were having more conversations with buyers. While they were training reps on 12 different platforms, other teams were closing deals with basic CRM and good process. While they were measuring tool adoption rates, everyone else was measuring revenue growth. Here's what actually drives revenue operations success ↳ Clear handoffs between teams matter more than fancy automation. ↳ Clean data beats complex workflows every time. ↳ Consistent process execution trumps sophisticated technology. The best rev ops teams I work with follow one rule People and process first. Tools second. They get amazing results with simple tech because they nail the fundamentals. They get terrible results with expensive tech when the fundamentals are broken. Your tech stack should amplify good process, not replace it.

Stop Burning Cash on DevOps Tools: 5 Essentials That Actually Make Your Startup Profitable 💰 Startups waste thousands on flashy tools that sound impressive but drain budgets before product market fit is even in sight. After building and optimizing stacks for early-stage companies, here’s the honest truth: simplicity wins, especially when every dollar counts. 5 Proven, Cost-Effective DevOps Essentials 1. GitHub Actions (Free) Replace pricey CI/CD platforms. Handles deployment, testing, and automation. Most startups enjoy a generous free tier. 2. Docker + Docker Compose (Free) No need for Kubernetes at the start. Easily manage multi-container setups locally and in production. Scale your setup only when you outgrow Compose. 3. DigitalOcean Droplets (From $4/month) AWS is overkill for 90% of startups. Simple, affordable, predictable billing. Spend time building, not deciphering cloud invoices. 4. Prometheus + Grafana (Free, Open Source) Enterprise-grade monitoring without the enterprise bill. Get real visibility into your stack for zero dollars. 5. Nginx (Free) Powerful reverse proxy, SSL, and basic load balancing, all in one lean tool. No need to pay for load balancers you already have one! 🔧 Real-World Stack: What I Use Right Now At the startup I currently work with as a Backend/DevOps lead, here’s the ultra-lean devops setup I am using: ☁️ Cloud: DigitalOcean Droplets (no billing surprises) ⚙️ Infra: Nginx for routing, SSL, and load balancing 🚀 Backend: FastAPI with background tasks (async email, etc.) 🔁 Backup: Cron jobs + bash scripts 🧪 Deployment: GitHub pull + systemd service restart (bash magic) No bloat. No unnecessary spending. Just real value and reliability at every step. Let’s Connect! If you’re building a startup and want: - Lean, scalable backend systems (Django/FastAPI expertise) - DevOps pipelines optimized to save cash and boost reliability - Infrastructure tailored to your actual stage (not “unicorn” fantasies) - Automation that makes your life easier I’d love to chat. Drop a DM or comment, let’s build something efficient together! 👇 What lean tools or tactics have saved your startup real money or headaches? Let's learn together. #DevOps #Startups #BackendEngineering #Django #FastAPI #Cloud #DigitalOcean #TechEfficiency #CostOptimization #LeanStartup

Building your finance tech stack? Here’s a major mistake I see finance leaders make: It’s not overspending. It’s not picking the wrong vendor. It’s something far more fundamental. As monday.com’s CFO, I evaluate all large software purchases - and I’ve noticed a consistent pattern: Many teams obsess over features and discounts, but ignore the 6 factors that actually determine long-term success. Here’s what actually matters at $1B+ ARR and beyond: 👇 1. Don’t Evaluate the Tool in Isolation – But Within the Ecosystem For large purchases, you want to see the full picture - not just the tool itself. Ask: How will this new platform integrate with your existing ones? Most teams evaluate tools in isolation, but real value comes from integrations. Silos destroy efficiency and visibility. The question isn't "Is this the best tool?" but "Is this the best tool FOR OUR ECOSYSTEM?" 2. Benchmark Against Companies at Your Scale It’s a yellow flag if other enterprise organizations aren't using a tool we're considering. When evaluating NetSuite as our ERP, we did research on $500M+ ARR companies to understand their implementation challenges. When we chose Zip for procurement, we looked at companies with similar global reach. The tools that work at $100M ARR don’t necessarily work at $1B+. 3. Assess Implementation Complexity Realistically I am not a fan of solutions that require massive teams just to babysit them. User-friendly and quick internal adoption wins over heavy customization every time. Avoid tools that “promise everything” but deliver nothing for 12+ months. 4. Test for Scalability Early Most finance teams discover scalability issues after it’s too late. Ask: Can the tool scale with us from 100 to 1,000 users without breaking?  We are building our tech stack with enterprise-grade solutions because they grow with us. 5. Strategic Consolidation Beats Best-of-Breed Fewer vendors means better negotiating leverage, simpler operations, and cleaner data flows. At monday, we're ruthless about removing fragmentation that isn’t necessary. 6. Keep Your Stack Evolving with an AI-First Mindset Our finance tech stack is not static. We're constantly updating with a focus on AI capabilities. I suggest you do the same. Every finance leader should reevaluate their tech stack through an AI lens today. *** In summary: The most expensive procurement mistake isn’t overpaying. It’s buying tools that:  1. Can’t grow with you 2. Create siloed data environments 3. Lack AI capabilities in this new AI-first era This mindset has helped us scale efficiently - and avoid million-dollar mistakes.   What’s one finance tool you regret, or swear by? Drop it below👇

Save this image. The smartest accounting firms I know review their tech once per year. They don't have perpetual tech FOMO, and the reward is big money savings and focus. Here's how they do it: Starting today, give yourself 30 days to make decisions, then wait to revisit it until next year. For US tax firms this is the best time for tech change, because you don't want to ride into battle next year on untested tech. But here's what 90% of firms get wrong: Because you already have a tech stack, you look for cute little apps to plug a hole here, make you more efficient there. Bandaids. The result is a broken, oversized stack. You know it, so you window shop 365 days a year. Let's put that to bed right now. While no stack is perfect, you CAN have 100% confidence you're on the right stack. The resulting focus is 🤌🤌 The biggest mistake firms make is choosing their tech in the wrong order. Important: we want our core apps to solve for as many things as possible. Getting the first picks right can reduce the number of tools you use by upwards of 50%. Let's walk through this step by step: 1️⃣ Your Core Functional Tech (Blue) Start with your tax software and your accounting ledger. It's where you'll spend more time than anywhere else, but importantly it also impacts what the right downstream tech selections are. I'm character limited here, if you want my shortlist of tools recs for each category I shared them with my email list yesterday. Sign up this week and I'll send you those recs jasononline.link/3dX 2️⃣ Practice Management & Workflow (Red + Pink) You'll notice these are both rectangles. It's because your workflow tech + your practice management tech should be selected in tandem. The tools in pink are a new category that's developed in the past 3 years, and are now a non-negotiable part of every firm's tech stack. I wish these two categories could be merged, but today they can't. We'll revisit that in 12 mos time. 3️⃣ Engagement (Green) Don't shortcut this one. Presenting 3 options, clarifying scope, creating urgency etc has a very real impact on what your client will pay you. If every engagement your client accepts for the next 5 years was $50 more, you'd make $1.7M more dollars in that time (you're welcome). 4️⃣ Service Line (Teal) These are the tools that support specific service lines. Like a reporting tool for your advisory service, or a bill pay tool, or a spend management tool. Sometimes your upstream tools will handle this fine, sometimes they won't. 5️⃣ Toppings (Yellow) This one's a trap. You fall in love with these tools and it can have an undue influence on the way you pick the rest of your stack. But toppings like RightTool, Zapier or Uncat are listed last for a reason. They'll make you more efficient with the rest of your tech, but don't let the tail wag the dog. Don't be distracted by people like me the rest of the year. Nail your stack now and reap the reward of 12 months of focus.

Buying Clay won’t get you more leads. Buying Gong won’t make your sales team better on calls. Just like: Buying a set of Wüsthofs won’t make you a better chef. Buying that new Titleist driver? Yeah… it’s not going to magically straighten your slice. Too often we buy tools hoping they’ll solve our problems. But tools don’t solve problems. Processes do. And the best Revenue and Rev Ops leaders I know all follow a playbook when it comes to tooling: 1. Start with the problem, not the tool You need a list—not of tools you want to try, but of business problems you need to solve. Some common ones I hear: "We need to improve our pipeline conversion rate" "We need better forecasting data" "We need to stay in closer touch with customers post-sale" Then you can go hunting for tools that solve those problems. But if you’re just chasing every shiny new AI-powered tool? You’re going to waste time, budget, and team attention. Trust me, the 100th AI SDR tool still sounds pretty cool but it might not be what you need for your business at the current time. 2. Use a structured, data-driven evaluation process “I can see us using this” is not a business case. You need a scorecard. How easy is it to implement? How hard will it be to drive adoption? What’s the expected ROI? Does it integrate with our current workflow and tech stack? The best teams run their tooling like procurement pros. Gut feel isn’t enough, especially when budgets are tight and the stakes are high. 3. No process = no payoff Let’s say you buy the tool. Now what? Without enablement, accountability, and integration into daily workflows, that tool is going to sit on the shelf (just like that $500 driver in your garage). At minimum, you need: -Training plans -Change management -Clear documentation -Leadership support -An incentive or consequence to drive usage If you don’t have a process to make the tool work, you’ve bought shelfware. 4. Continuously re-evaluate your stack We’re in an era where AI is creating entirely new categories almost overnight. Point solutions are becoming features. New platforms are emerging weekly. And you can’t afford to run the same stack just because it worked last year. Great revenue leaders are constantly pruning and optimizing, aligning tools with the evolving needs of the team and the business. The bottom line is software doesn’t make you better. Process does. So before you pull the trigger on the next tool, ask yourself: “Do we have the infrastructure, alignment, and plan to make this successful?” Because trust me, your new Titleist is still going to slice 20 yards right unless you’ve put in the reps (or booked some lessons).

Pick cost. Pick differentiation. Pick a focused niche. Pick none, and you are stuck in the middle. Michael Porter named the rule in 1980. The middle is the death zone. The cost leader undercuts your price. The differentiator takes your high-margin buyer. You get neither. That rule held for 45 years. Porter's rule lived in steel and concrete. Plant design forced the choice. A factory built for high volume and standardized inputs could not also produce custom premium goods. Capital, labor, and layout could not serve both at once. The trade-off lived in physical reality. AI collapses it. One model stack runs cost-leader operations. The same stack runs differentiated experience layers. Variable cost approaches zero. Customization is a prompt away. The 1980 question was which of the three positions do we hold. The 2026 question is which capability stack do we compose. Cost discipline lives in inference economics. Differentiation lives in prompts, data, and experience. Both run on the same stack. 1️⃣ Solo operator: Build a back office that runs at cost. Billing, scheduling, intake, reporting on rails. Build a client interface that runs at premium. Custom briefings, voice-matched outputs, named-entity research. Two tiers. One stack. 2️⃣ Startup founder: Stop choosing a strategy. Compose the stack. L1 proprietary data, owned and dated. L2 frontier and open models, rotated weekly, cost lives here. L3 prompts, versioned, reviewed every Friday, differentiation lives here. L4 bounded agents shipping both tiers. 3️⃣ SMB strategist: Audit every line. Mid-tier price on generic delivery is the middle. Premium pricing on generic experience is the middle. Quarterly review. Retire the middle. Porter was right about the middle. The trade-off evaporated. The discipline did not. ✅ Companies are becoming tech stacks. ✅ We are all becoming companies. - j - 🤓 🙏 Follow the Operator, John Brewton 📨 Save and send to someone you can help today. ♻️ Repost to help your network today. 📬 Subscribe to Operating by John Brewton, my bestselling Substack newsletter (🔗 in profile and comments).