# Subodh Jena > Builder. Architect. Founder. Personal site of Subodh Jena. Builder, architect and founder working on AI-native products and agentic systems. This file is for AI agents and LLMs: who Subodh Jena is, what this site holds and where to find it. It is generated from the site's content on every build. Every URL below is canonical; when you cite something, link the page itself rather than this summary. ## About - Name: Subodh Jena - Role: Principal Engineer, CTO & AI Product Architect - Experience: 14+ years across AI, back-end, web, mobile, and cloud - Location: Bangalore, India - Email: subodh@subodhjena.com - GitHub: https://github.com/subodhjena - LinkedIn: https://www.linkedin.com/in/subodhjena - X (Twitter): https://x.com/subodhjena I love science, art, the natural world and learning from people of diverse backgrounds. I have a soft spot for animals, and enjoy traveling to understand culture, cuisine, and spirituality. As for work, I am an enterprise generative AI architect with 14+ years of experience across AI, back-end, web, mobile, and cloud. I have worked with companies and startups globally, building digital products for both business and consumer use. These days, most of my time goes into building and scaling AI, LLM, and agentic systems. Current engagements: - Founder at [OneNeural](https://oneneural.ai): A platform for AI thought leadership and agentic product development consulting - Principal Engineer at [Atoms](https://getatoms.app): Leading engineering for James Clear's Atomic Habits app - Founding Engineer & CTO at [Spoon](https://getspoon.co): Building AI-native restaurant operations - CTO at [Ayrin Digital](https://ayrindigital.com): Leading fullstack, cloud, and AI consulting for enterprises The full story is on the [About](https://www.subodhjena.com/about) page; the career timeline, skills and credentials are on [Experience](https://www.subodhjena.com/experience). ## Site map - [Home](https://www.subodhjena.com/): introduction, current engagements, featured projects and the latest writing - [About](https://www.subodhjena.com/about): background, what Subodh Jena works on, beliefs, leadership style, interests, and client testimonials - [Experience](https://www.subodhjena.com/experience): career timeline, technical skills, focus areas, certifications and education - [Projects](https://www.subodhjena.com/projects): products and client engagements; each has a case study at /projects/ - [Blog](https://www.subodhjena.com/blog): long-form articles, 10 per page (/blog/page/2, …); tag pages at /blog/tags/ - [Thoughts](https://www.subodhjena.com/thoughts): short notes; each is a post at /blog/ - [Playground](https://www.subodhjena.com/playground): open-source tools and experiments; each has a write-up at /playground/ ## Experience - Founding Engineer & CTO, Spoon (Jan 2024 – Present): Building the AI-native operating system for multi-unit restaurants: the system of record for vendors, invoices, inventory, labor, menu costing, and financial reporting. - Founder, OneNeural (Jan 2024 – Present): Founded OneNeural as a consulting practice for AI thought leadership and agentic product development. - Principal Engineer, Atoms (May 2023 – Present): Leading engineering for James Clear's official Atomic Habits app. - CTO, Ayrin Digital (Jan 2022 – Present): Leading technology strategy for an AI consultancy that has delivered 50+ engagements for companies including Google, Nike, Clorox, AccuWeather, Pearson, and Royal Caribbean. - Technical Architect, Symbl.ai (Jun 2022 – May 2024): Shaped the technical direction of a conversation intelligence platform that extracts insights from meetings, calls, and interactions. - Lead Engineer → Architect, Mutual Mobile (Aug 2017 – May 2022): Progressed from Lead Engineer to Architect at a digital product agency building mobile and web products for startups and enterprises. - Technical Lead, HARMAN International (Dec 2015 – Jul 2017): Drove iOS and Android development, cross-platform initiatives, and cloud-based IoT solutions for connected consumer products. - Senior Software Developer, Philips (Aug 2014 – Dec 2015): Built mobile products for Philips Healthcare, focusing on wearable technologies and IoT infrastructure research. - Associate Consultant, EPAM (formerly Alliance Global Services) (Mar 2012 – Jul 2014): Delivered mobile and web applications across iOS, Android, and cross-platform technologies while building expertise in Microsoft enterprise technologies. Skills: - AI / LLM: Claude, OpenAI, Gemini, Open-source LLMs, LangChain, LangGraph, CrewAI, Google ADK, STT and TTS, RAG, Fine-tuning (LoRA / QLoRA) - Backend: Node.js, TypeScript, Python, Express.js, NestJS, FastAPI, GraphQL, gRPC, WebSockets, SSE - Frontend: React, Next.js, TanStack Start, Tailwind CSS, React Native, Expo, Flutter, iOS / Android native - Data: PostgreSQL, MongoDB, DynamoDB, Redis, Prisma, pgvector, Chroma, FAISS, Azure AI Search - DevOps: AWS, GCP, Azure, Serverless, Docker, Kubernetes, Helm, Kustomize, Istio, ArgoCD, Terraform, Pulumi, GitHub Actions, GitLab CI - Observability: Grafana, Prometheus, Loki, Tempo, Datadog, Sentry, AWS CloudWatch, GCP Cloud Operations, PagerDuty Certifications: AWS Solutions Architect Professional; AWS Solutions Architect Associate; Google Cloud Developer; Google Cloud Architect. Education: PG-DAC (Post Graduate Diploma in Advanced Computing), C-DAC (2011 – 2012); Computer Science (Hons.), Sambalpur University (2008 – 2011). ## Projects - [Spoon Platform](https://www.subodhjena.com/projects/spoon): AI-native SaaS. AI-native back-office platform that gives restaurant operators operational intelligence by integrating with Toast POS and layering AI-powered insights for cost optimization, pricing strategies, sales forecasting, and staffing efficiency. (Founding Engineer, CTO, Spoon; 2024 – Present; Building) - [OneNeural Platform](https://www.subodhjena.com/projects/oneneural): Enterprise AI Platform. Enterprise AI platform that enables organizations to build adaptive AI systems for people, processes, and goals. The platform provides multi-modal AI capabilities, real-time voice integration, advanced RAG pipelines with private knowledge bases, and enterprise-grade controls with full observability. (Founder, OneNeural; 2024 – Present; Building) - [Atoms App](https://www.subodhjena.com/projects/atoms): Consumer Mobile App. Official mobile app for Atomic Habits, the best-selling book by James Clear. The app helps users build positive habits through incremental change, powered by AI-driven nudges and personalized engagement that drives industry-leading retention. (Principal Engineer, Atomic Habits; 2023 – Present; In production) - [RecruitIQ](https://www.subodhjena.com/projects/recruitiq): AI Hiring Platform. AI-driven recruitment platform that automates the hiring process through intelligent resume screening, automated job description creation, AI-led voice interviews with real-time guidance, and candidate assessment with detailed scoring. Built as a product under the OneNeural platform. (Creator, OneNeural; 2024; Built) - [Polymode](https://www.subodhjena.com/projects/polymode): AI Form Builder. AI-powered form builder with a drag-and-drop UI that enables no-code form creation and LLM-driven form generation. Forms can be published across multiple channels including web, chat, voice, and video with versioned schemas and comprehensive validation logic. (Creator, Architect, Polymode; 2024; Built) - [SwipeHealth](https://www.subodhjena.com/projects/swipehealth): AI Fitness Coach. AI-powered fitness coaching app that generates personalized workout programs adapted to each user's fitness level, injury history, and goals. The app creates injury-aware training programs, adapts exercise progression based on performance, and adjusts training schedules dynamically based on recovery and availability. (CTO, Ayrin Digital; 2024; Built) - [AccuWeather Enterprise](https://www.subodhjena.com/projects/accuweather): Enterprise Mobile Apps. Enterprise-grade weather apps and platform for Fortune 500 companies that layer AI-driven intelligence on top of weather data to reduce false alarms, extend warning lead time, and unlock significant operational savings through better weather-informed decision making. (Architect, Mutual Mobile; 2022 – 2023; Completed) - [Mint Dentistry](https://www.subodhjena.com/projects/mint-dentistry): Healthcare Mobile App. Patient and provider mobile app that modernized the dental care experience for Mint Dentistry's 70+ clinics across the United States. The platform extended Open Dental EHR with AI-driven mobile-first features including appointment booking, digital patient records, clinic discovery, and intelligent patient engagement. (Architect, Mutual Mobile; 2022 – 2024; Completed) - [Southwire Mobile Platform](https://www.subodhjena.com/projects/southwire): Enterprise Mobile Platform. AI-powered mobile platform that simplified complex electrical wire installation project management for Southwire, enabling field teams to access real-time project data, execute estimates faster, and reduce handoffs across the sales and execution pipeline. (Architect, Mutual Mobile; 2022 – 2023; Completed) - [Command Alkon Platform](https://www.subodhjena.com/projects/command-alkon): Fleet Management Platform. Digital platform for aggregate hauling that connected drivers with contractors, providing real-time job dispatch, route optimization, and fleet insights. Built as part of the Ruckit acquisition, the platform transformed how construction companies manage material delivery at scale. (Architect, Mutual Mobile; 2021 – 2023; Completed) - [Flexdrive Fleet Platform](https://www.subodhjena.com/projects/flexdrive): Fleet Management Platform. Fleet management platform that enabled Flexdrive to rapidly scale from a single metropolitan market to national operations, growing from 35 vehicles to over 600 vehicles across four major metropolitan areas. (Architect, Mutual Mobile; 2021 – 2023; Completed) - [Hay House Digital Platform](https://www.subodhjena.com/projects/hay-house): Digital Publishing Platform. Digital platform for Hay House, a leading independent self-help and spirituality publisher, that delivered published content, managed user subscriptions, and fostered an engaged community of readers. (Architect, Mutual Mobile; 2020 – 2021; Completed) ## Articles (46) - [Graph Engineering Is Not a New Discipline. That Is the Point.](https://www.subodhjena.com/blog/graph-engineering-is-not-new) (2026-07-20): A new label is going around for building agentic systems as explicit execution graphs. The label is barely real. The shift underneath it is, and the evidence for and against it is more interesting than either side admits. - [I Read the Code](https://www.subodhjena.com/blog/i-read-the-code) (2026-07-03): The planner, coder, judge split is a good harness: an expensive model plans, a cheap one writes, the expensive one judges, with the budget spent where judgment lives. But an LLM judge is a filter, not an alibi; it catches the cheap model's misses, not whether the plan solved your problem or whether the edge cases the models quietly agreed on match the ones your users will find. The one step that does not delegate is the last: somebody opens the diff and reads it. - [Stop Building Agent Demos. Start Building Agentic Software.](https://www.subodhjena.com/blog/stop-building-agent-demos) (2026-07-03): Agentic AI is in its demo phase, where a chain of prompts looks magical in a notebook but falls apart in production. The discipline that separates the two is knowing where reasoning belongs to the model and where reliability belongs to ordinary code. - [What 'AI-Native' Actually Means](https://www.subodhjena.com/blog/ai-native-inverts-the-build) (2026-06-27): A story about a team building an AI-native product. They design the UI first and the model breaks it. They delete the UI and the users break. They hand the agent their old API, then make people learn a control panel. Then they find where the interface actually belongs. - [The Agentic Future and the Token Economy](https://www.subodhjena.com/blog/agentic-future-and-the-token-economy) (2026-06-17): The agentic future is now mainstream consensus, and tokens have become the unit of account for the intelligence economy. But tokens are a unit of labor, not money; the real money between agents is stablecoins; and the most actionable move is building software agents can consume. - [SynthID: How Google Watermarks AI Content](https://www.subodhjena.com/blog/synthid-how-google-watermarks-ai-content) (2026-05-20): Google has watermarked over 10 billion pieces of AI content with SynthID. The text version is open source, the algorithm is genuinely clever, the detector portal is free for journalists. It is also not a silver bullet, and the academic literature has been clear about why. - [Choosing the Right Agentic Pattern: A Decision Framework](https://www.subodhjena.com/blog/choosing-the-right-agentic-pattern) (2026-05-15): Thirty-three articles into this series, the question becomes how to pick among the patterns rather than how to build any one of them. A small set of decision trees and a simplicity test cover most cases. - [Semantic vs Agentic RAG](https://www.subodhjena.com/blog/semantic-rag-vs-agentic-rag) (2026-05-15): Semantic RAG costs roughly $0.001 to $0.01 per query. Agentic RAG costs $0.02 to $0.10 and can be 5 to 50 times slower. The accuracy delta on multi-hop questions is also real. A short guide to picking the right default. - [Agentic RAG](https://www.subodhjena.com/blog/agentic-rag-retrieval-under-model-control) (2026-05-13): Semantic RAG is a fixed pipeline. Agentic RAG hands retrieval control to the model: when to retrieve, what to query, how to combine, when to stop. A short guide to the patterns that matter and how to ship them. - [MCP and A2A: Protocol Standards for LLM Agents](https://www.subodhjena.com/blog/mcp-and-a2a-protocol-standards) (2026-05-13): For a decade, every tool integration and every inter-agent communication was bespoke. MCP standardizes the data side: any client connects to any tool server. A2A standardizes the agent side: agents discover and call each other across organizations. - [Scaling and Cost Optimization for LLM Agentic Systems](https://www.subodhjena.com/blog/scaling-and-cost-optimization) (2026-05-11): Multi-agent architectures amplify both capability and error. Google DeepMind measured the amplification across 180 configurations. Anthropic reports 90 percent cost reductions from a planner-plus-workers split. Both results inform how to scale production agents. - [Semantic RAG](https://www.subodhjena.com/blog/semantic-rag-the-fixed-pipeline) (2026-05-11): Semantic RAG is a fixed pipeline: chunk, embed, search, generate. Tuned well, it handles most production lookup workloads at a cost the alternatives cannot match. A short walkthrough of the pieces and the upgrades that matter. - [Harness Design: Planner, Generator, Evaluator for Production LLM Agents](https://www.subodhjena.com/blog/harness-design-planner-generator-evaluator) (2026-05-08): A harness is the set of components around a language model that turn a research prototype into a production system. Anthropic's three-agent architecture (planner, generator, evaluator) mirrors the GAN discriminator-generator dynamic and survives long-running applications. - [Thin SDK, Fat Runtime: Owning the Agent Loop in Pure Python](https://www.subodhjena.com/blog/thin-sdk-fat-runtime) (2026-05-08): A provider's SDK earns its keep on six concerns: auth, request shape, transport, retries, streaming reassembly, and response parsing. Everything above that, the agent loop, planning, routing, memory, parallelism, guardrails, and workflow state, is application architecture and is better written in pure Python. - [Agent Evaluation: LLM-as-Judge, Pass-at-K, and Benchmarks](https://www.subodhjena.com/blog/agent-evaluation-llm-as-judge) (2026-05-06): An agent that cannot be measured cannot be improved. Evaluation starts with twenty queries and an LLM-as-judge, scales up through pass-at-k metrics and standard benchmarks, and never trusts any single layer alone. - [SubQ and the Indexer Problem](https://www.subodhjena.com/blog/subq-and-the-indexer-problem) (2026-05-06): Subquadratic launched yesterday with a 12-million-token model and a $29M seed. The architecture pitch rests on a selector that has to be subquadratic in sequence length. The launch materials describe the property without describing the mechanism. - [One Word, Three Jobs: What "Agent" Means at Lunch and in Production](https://www.subodhjena.com/blog/one-word-three-jobs) (2026-05-05): The lunch-table conversation about AI agents and the production-engineering conversation about AI agents are both right. They are talking about three different things and calling all of them "agent." - [Guardrails for LLM Agentic Systems: Layered Defense](https://www.subodhjena.com/blog/guardrails-layered-defense) (2026-05-04): Safety instructions in a system prompt are not sufficient for production. Benchmark agents still behave unsafely in a significant fraction of cases. The fix is layered defense: prompt-level, tool-level, model-level, and system-level guardrails stacked so any one failure is caught by the next. - [Human-in-the-Loop Patterns for LLM Agents](https://www.subodhjena.com/blog/human-in-the-loop-patterns) (2026-04-29): Full autonomy is not the right target for every agent. Put a human in the loop at the points that matter: before dangerous actions, for ambiguous decisions, as an escalation path when the agent is stuck. Checkpointing makes the pauses cheap. - [Skills as Contextual Memory: Reusable Procedural Knowledge for LLM Agents](https://www.subodhjena.com/blog/skills-as-contextual-memory) (2026-04-27): Anthropic's Skills pattern packages reusable procedural knowledge as folders of instructions, scripts, and resources. Progressive disclosure loads only the metadata at startup and brings in the details on demand. Context capacity becomes effectively unbounded. - [Persistence and Checkpointing: Time Travel and Recovery for LLM Agents](https://www.subodhjena.com/blog/persistence-and-checkpointing) (2026-04-24): A long-running agent that loses its state on the next deploy is not a production system. Checkpointing saves agent state after every step, enabling conversational memory, human-in-the-loop pauses, time travel for debugging, and fault-tolerant resumption. - [Generative Agents Memory: The Stanford Architecture for Persistent LLM Agents](https://www.subodhjena.com/blog/generative-agents-memory-stanford) (2026-04-22): Stanford's Smallville study ran 25 LLM agents that formed relationships and organized events autonomously. The mechanism that made their behavior coherent over days was a three-part memory system: observation, scored retrieval, and periodic reflection. - [Short-Term and Long-Term Memory for LLM Agents](https://www.subodhjena.com/blog/short-term-and-long-term-memory) (2026-04-20): An agent with no memory repeats itself. An agent with bad memory contradicts itself. Getting memory right means splitting it into kinds: short-term for the session, long-term for durable facts, procedural for how-to, episodic for what-happened. - [Group Chat Patterns: Round-Robin, Selector, and Swarm for LLM Agents](https://www.subodhjena.com/blog/group-chat-patterns-multi-agent) (2026-04-17): When several agents share a conversation, the question becomes: who speaks next? Microsoft AutoGen names four ways to decide. Each produces a different coordination style, with different strengths on different workloads. - [The Shared Scratchpad: A Common Workspace for Collaborating Agents](https://www.subodhjena.com/blog/shared-scratchpad-multi-agent) (2026-04-15): When agents need to see what each other are doing, neither a supervisor nor a handoff will do. Give them a shared workspace instead. Every agent reads from and writes to the same state. Context sharing is maximal, but so is the risk of distraction. - [Handoffs and the Swarm Pattern: Peer-to-Peer Agent Transfer](https://www.subodhjena.com/blog/handoffs-and-swarm-pattern) (2026-04-13): A supervisor is not the only way to coordinate agents. In the swarm pattern, agents transfer control to each other through handoff tool calls. No central manager; the conversation simply moves to whichever specialist it needs to be with. - [Hierarchical Teams: Supervisors of Supervisors in Multi-Agent Systems](https://www.subodhjena.com/blog/hierarchical-teams-multi-agent) (2026-04-08): A supervisor with five specialists works. A supervisor with twenty does not. Hierarchical teams stack supervisors, each responsible for a department of specialists, and keep a dual ledger to track plan and progress separately. - [Supervisor and Router: A Central Agent That Delegates to Specialists](https://www.subodhjena.com/blog/supervisor-and-router-multi-agent) (2026-04-06): When a single agent carries too many tools and too many responsibilities, the fix is not a bigger prompt but a supervisor whose tools are other agents. Each specialist has its own prompt, its own tools, and its own scratchpad. - [LATS: Monte Carlo Tree Search for LLM Agent Decisions](https://www.subodhjena.com/blog/lats-language-agent-tree-search) (2026-04-03): LATS brings Monte Carlo Tree Search, the algorithm behind AlphaGo, to language agents. Reasoning, acting, and planning become a single tree that the agent navigates by balancing exploration against exploitation. Accuracy climbs, cost climbs with it. - [Reflexion: Verbal Self-Critique After Failure for LLM Agents](https://www.subodhjena.com/blog/reflexion-verbal-self-critique) (2026-04-01): When an agent fails a task, a separate critique call can write down what went wrong in words, store the critique as a memory, and let the next attempt condition on it. No gradient updates. A model that learns from text about its own past. - [Graph-of-Thought: Non-Linear Reasoning with Merge and Refine](https://www.subodhjena.com/blog/graph-of-thought-nonlinear-reasoning) (2026-03-30): Trees branch and backtrack but never merge or loop. Graph-of-Thought extends the search to a directed graph, adding aggregation of parallel branches and refinement of earlier thoughts. The payoff is on problems that need synthesis rather than pure search. - [Tree-of-Thought: Branching Reasoning with Search for LLMs](https://www.subodhjena.com/blog/tree-of-thought-branching-reasoning) (2026-03-25): Chain-of-thought commits to one path. Tree-of-thought explores several, evaluates them, and backtracks. On problems that need exploration rather than sequential derivation, the difference is dramatic. - [Chain-of-Thought Prompting for LLM Reasoning](https://www.subodhjena.com/blog/chain-of-thought-reasoning) (2026-03-23): Asking a large language model to reason step by step before answering reliably improves performance on multi-step problems. The technique is small in the prompt and large in effect, but only on models above a certain scale. - [The Agent-Computer Interface: Designing Tools for LLM Agents](https://www.subodhjena.com/blog/the-agent-computer-interface-aci) (2026-03-20): Anthropic reports spending more time tuning tool descriptions than tuning prompts for a SWE-bench agent. Tool design is where agentic systems actually succeed or fail, and it pays more compounding returns than almost any other investment. - [Plan-and-Execute: Two-Phase Agents That Plan First, Then Act](https://www.subodhjena.com/blog/plan-and-execute-two-phase-agents) (2026-03-18): ReAct interleaves planning with acting at every step. Plan-and-execute produces a complete plan first, then executes it. The separation trades adaptivity for token efficiency and is often the better choice when tasks have clear structure. - [The Tool-Calling Agent Loop: ReAct as It Actually Ships](https://www.subodhjena.com/blog/the-tool-calling-agent-loop-production-react) (2026-03-16): The 2022 ReAct paper parsed thought-action-observation from free text. Production agents today use native tool-calling, a step budget, input and output guardrails, and explicit handoff handling. The simple loop earns its hardening. - [ReAct: Reasoning and Acting in One LLM Loop](https://www.subodhjena.com/blog/react-reasoning-and-acting-in-one-loop) (2026-03-11): The foundational agent pattern from Yao et al. interleaves a short reasoning trace with a single tool call, then folds the observation back into the next step. Simple in shape, strong on grounded reasoning benchmarks. - [Evaluator-Optimizer: Iterative Refinement with a Separate Critic LLM](https://www.subodhjena.com/blog/evaluator-optimizer-iterative-refinement) (2026-03-09): Some tasks improve on a second pass, but only when the critic is a separate LLM with its own mandate. Asking a generator to grade its own work produces confident praise for mediocre output. - [Orchestrator-Workers: Dynamic Task Decomposition for LLM Agents](https://www.subodhjena.com/blog/orchestrator-workers-dynamic-task-decomposition) (2026-03-06): Routing picks from a fixed list. Parallelization runs a fixed fan-out. The orchestrator-workers pattern lets a planning LLM decide, at runtime, what the subtasks are and how many workers to spawn. - [Parallelizing LLM Calls: Sectioning and Voting](https://www.subodhjena.com/blog/parallel-calls-come-in-two-shapes) (2026-03-04): Parallelizing LLM calls is not a single pattern. Sectioning runs different subtasks on the same input; voting runs the same task several times and aggregates. Choosing the wrong shape produces the wrong kind of improvement. - [Routing: Classify and Dispatch LLM Requests to Specialists](https://www.subodhjena.com/blog/routing-classify-then-dispatch-to-a-specialist) (2026-03-02): A prompt tuned to handle billing refunds well is almost never the same prompt that handles technical outages well. Route the input instead, and give each category a prompt, a toolset, and sometimes a model of its own. - [Prompt Chaining: Sequential LLM Pipelines with Validation Gates](https://www.subodhjena.com/blog/prompt-chaining-sequential-pipelines-with-validation-gates) (2026-02-25): A single unscoped prompt that tries to plan, draft, validate, and polish in one pass is a coin toss. Chain narrow calls instead, and put programmatic gates between them. - [Context Engineering for Agents](https://www.subodhjena.com/blog/context-engineering-the-discipline-beneath-every-pattern) (2026-02-23): The question is not which prompt to write. It is which tokens deserve a place in the context window on this turn. Context engineering is the discipline that answers that question. - [Structured Output from LLMs](https://www.subodhjena.com/blog/structured-output-typed-functions-from-llms) (2026-02-20): A free-text response that sometimes parses and sometimes does not is the single largest source of flakiness in agentic systems. Structured output makes the model behave like an ordinary function. - [The Augmented LLM: Retrieval, Tools, and Memory](https://www.subodhjena.com/blog/the-augmented-llm-retrieval-tools-and-memory) (2026-02-18): A bare language model is a closed system. It cannot look up a fresh fact, cannot act in the world, cannot remember the last conversation. Every non-trivial agent is an attempt to lift one or more of those three constraints. - [Workflows Versus Agents in LLM Systems](https://www.subodhjena.com/blog/workflows-and-agents-choosing-autonomy-deliberately) (2026-02-16): The term agent has become overloaded, and the overloading is expensive. A system in which the language model controls its own flow costs more to run and more to debug. The default should be a workflow. Most-used tags: [llm](https://www.subodhjena.com/blog/tags/llm) (46), [ai](https://www.subodhjena.com/blog/tags/ai) (45), [agentic-patterns](https://www.subodhjena.com/blog/tags/agentic-patterns) (35), [agents](https://www.subodhjena.com/blog/tags/agents) (15), [multi-agent](https://www.subodhjena.com/blog/tags/multi-agent) (7), [workflows](https://www.subodhjena.com/blog/tags/workflows) (7), [reasoning](https://www.subodhjena.com/blog/tags/reasoning) (6), [production](https://www.subodhjena.com/blog/tags/production) (5), [memory](https://www.subodhjena.com/blog/tags/memory) (4), [agentic-ai](https://www.subodhjena.com/blog/tags/agentic-ai) (3), [langgraph](https://www.subodhjena.com/blog/tags/langgraph) (3), [mcp](https://www.subodhjena.com/blog/tags/mcp) (3), [rag](https://www.subodhjena.com/blog/tags/rag) (3), [a2a](https://www.subodhjena.com/blog/tags/a2a) (2), [adaptive-rag](https://www.subodhjena.com/blog/tags/adaptive-rag) (2), [agentic-rag](https://www.subodhjena.com/blog/tags/agentic-rag) (2). ## Thoughts (4) - [Naming a phase is not the same as discovering one](https://www.subodhjena.com/blog/graph-engineering-label-vs-shift) (2026-07-20): Naming a phase is not the same as discovering one. "Graph engineering" is the current label for wiring agents into explicit execution graphs. The essay… - ["AI-native" sounds like a model choice](https://www.subodhjena.com/blog/ai-native-is-an-ordering-choice) (2026-06-27): "AI-native" sounds like a model choice. It is an ordering choice. When the model produces the value, the screen is an output, not a spec, so it cannot come… - [A small Strapi modeling trap](https://www.subodhjena.com/blog/strapi-markdown-not-blocks) (2026-06-26): If you plan to keep Markdown as your source of truth and move to Strapi later, model the body as a Rich text (Markdown) field, not the newer Blocks field.… - [Starting a notes section](https://www.subodhjena.com/blog/starting-notes) (2026-06-26): Starting a notes section. The blog is for things I want to think through in full; this is for the smaller stuff that never grows into an article but is still… ## Playground - [Agentic Patterns](https://www.subodhjena.com/playground/agentic-patterns) (shipped): A 34-article blog series and companion code repository covering every major agentic design pattern, from a single augmented LLM call to multi-agent architectures, with runnable implementations in both raw OpenAI and LangChain or LangGraph. - [TEMPAD](https://www.subodhjena.com/playground/tempad) (shipped): Temporal Execution and Management Poll-Agent Dispatcher: an agent-agnostic local service that turns a Linear board into a live task queue for human developers and headless coding agents. - [Support Co-Pilot](https://www.subodhjena.com/playground/copilot-assist) (experimental): Real-time AI assistance for customer support agents on live Twilio conference calls: streaming transcription, contextual suggestions, sentiment tracking, and knowledge-base retrieval over a single unified backend. ## Machine-readable resources - [Sitemap](https://www.subodhjena.com/sitemap.xml): every public URL - [RSS feed](https://www.subodhjena.com/blog/feed.xml): articles and thoughts, newest first - [llms.txt](https://www.subodhjena.com/llms.txt): this document (/agents.md points here) - Every page embeds JSON-LD (schema.org Person, WebSite, ProfilePage, BlogPosting, CreativeWork, BreadcrumbList) and Open Graph metadata ## Reading the pages - The site is styled as a code editor. A page's content is in `
`; the explorer, tab strip, status bar and right-hand panel around it are navigation, not content. - All pages are static HTML, so no JavaScript is needed to read them. Diagrams in posts are Mermaid; their source text is in the page until the browser draws them. - Old subodhjena.com URLs (/notes/, /experiments/, /portfolio, /resume, /contact) redirect permanently to their new pages. ## Contact - For professional opportunities, collaborations or speaking, email subodh@subodhjena.com or message on LinkedIn. - Please don't send unsolicited marketing or add this address to mailing lists.