Liliputtech.io explores AI, software engineering, cloud architecture, and emerging technologies through practical guides, real-world experiments, and honest technical analysis.https://liliputtech.io/https://liliputtech.io/posts/agent-success-coding/https://liliputtech.io/posts/agent-success-coding/GitHub Copilot’s coding agent opens a draft pull request and asks for review. The merge is still yours. Stripe’s Claude Code rollout is a vendor story with the same lesson.Sun, 04 Oct 2026 00:00:00 GMThttps://liliputtech.io/posts/agent-success-assisted/https://liliputtech.io/posts/agent-success-assisted/Cineplex’s refund copilot writes after a Guest Services agent types a booking ID. That is automation with a start button, not an unsupervised agent.Sat, 03 Oct 2026 00:00:00 GMThttps://liliputtech.io/posts/agent-success-support/https://liliputtech.io/posts/agent-success-support/Klarna’s 2024 assistant numbers, then the 2025 decision to hire humans again. Vendor stories keep their own dated figures — not a roundup article’s 70%.Fri, 02 Oct 2026 00:00:00 GMThttps://liliputtech.io/posts/agent-success-stories/https://liliputtech.io/posts/agent-success-stories/A number with a date is not a full story. Three lessons: support that can take action, operations that still need a person to start, and coding agents that wait for review.Thu, 01 Oct 2026 00:00:00 GMThttps://liliputtech.io/posts/multi-agent-where-it-shows/https://liliputtech.io/posts/multi-agent-where-it-shows/Five jobs where a graph is worth using: fraud rings, outage traces, supply-chain ripples, multi-step research, and non-linear agent handoffs.Wed, 09 Sep 2026 00:00:00 GMThttps://liliputtech.io/posts/multi-agent-when-not-to-split/https://liliputtech.io/posts/multi-agent-when-not-to-split/Start with one agent. Split only when the instructions, tools, or rules change. A merge that only counts agreement has not checked anything.Tue, 08 Sep 2026 00:00:00 GMThttps://liliputtech.io/posts/multi-agent-fan-out/https://liliputtech.io/posts/multi-agent-fan-out/Parallel specialists are a fan-out. The merge must see every branch’s write. LangGraph calls this pattern fan-out and fan-in.Mon, 07 Sep 2026 00:00:00 GMThttps://liliputtech.io/posts/multi-agent-the-map/https://liliputtech.io/posts/multi-agent-the-map/A node is a specialist agent with a loop and a stop. An edge says which agent goes next. State is what the next node is allowed to see.Sun, 06 Sep 2026 00:00:00 GMThttps://liliputtech.io/posts/multi-agent-the-graph/https://liliputtech.io/posts/multi-agent-the-graph/A four-part path: the map, fan-out then merge, when not to split, and where this structure shows up. Adding more AI personas does not make a system.Sat, 05 Sep 2026 00:00:00 GMThttps://liliputtech.io/posts/why-agents-fail-intelligence/https://liliputtech.io/posts/why-agents-fail-intelligence/A capable agent knows when to act, verify, stop, and ask — and the runtime stack that makes that judgement enforceable.Sun, 16 Aug 2026 00:00:00 GMThttps://liliputtech.io/posts/why-agents-fail-humans/https://liliputtech.io/posts/why-agents-fail-humans/Automation bias, multi-product consensus trust, Starbucks Korea’s Tank Day, and ambiguous responsibility as human failure modes.Sat, 15 Aug 2026 00:00:00 GMThttps://liliputtech.io/posts/why-agents-fail-benchmarks/https://liliputtech.io/posts/why-agents-fail-benchmarks/SWE-bench, OSWorld, METR, ReliabilityBench, and OmnilingualGAIA2 measure different things — one English score is not reliability.Thu, 13 Aug 2026 00:00:00 GMThttps://liliputtech.io/posts/why-agents-fail-seven-modes/https://liliputtech.io/posts/why-agents-fail-seven-modes/A non-exhaustive checklist of trajectory failure modes, plus why small per-step error becomes large end-to-end risk.Wed, 12 Aug 2026 00:00:00 GMThttps://liliputtech.io/posts/why-agents-fail-illusion/https://liliputtech.io/posts/why-agents-fail-illusion/Why impressive reasoning demos create unrealistic expectations for agents that must finish long workflows.Tue, 11 Aug 2026 00:00:00 GMThttps://liliputtech.io/posts/why-agents-fail/https://liliputtech.io/posts/why-agents-fail/What happens when an AI that is often right is given permission to act many times? A five-part series on capability vs reliability.Mon, 10 Aug 2026 00:00:00 GMThttps://liliputtech.io/posts/rag-vs-agents-automation/https://liliputtech.io/posts/rag-vs-agents-automation/When to use traditional automation, RAG, an LLM, or an AI agent — and when you should NOT use an agent.Thu, 06 Aug 2026 00:00:00 GMThttps://liliputtech.io/posts/rag-product-instances/https://liliputtech.io/posts/rag-product-instances/Named RAG product instances: Google NotebookLM and enterprise search patterns — instances of retrieve-then-generate, not a comparison of vendors.Wed, 05 Aug 2026 00:00:00 GMThttps://liliputtech.io/posts/rag-enterprise-ops/https://liliputtech.io/posts/rag-enterprise-ops/Enterprise RAG in production: corpus permissions, chunking, index freshness, and when retrieval returns the wrong chunk.Tue, 04 Aug 2026 00:00:00 GMThttps://liliputtech.io/posts/rag-architecture/https://liliputtech.io/posts/rag-architecture/RAG architecture in operations: index, retrieve k, generate. Lewis et al. pattern with diagrams and an internal wiki Q&A example.Mon, 03 Aug 2026 00:00:00 GMThttps://liliputtech.io/posts/rag-why-your-data/https://liliputtech.io/posts/rag-why-your-data/Why language models need your corpus: training cutoff, hallucination, and the enterprise shift from general AI to grounded retrieval.Sun, 02 Aug 2026 00:00:00 GMThttps://liliputtech.io/posts/rag-explained/https://liliputtech.io/posts/rag-explained/Why AI needs access to your data: a five-part path from general models to practical enterprise RAG, product instances, and choosing the right tool.Sat, 01 Aug 2026 00:00:00 GMThttps://liliputtech.io/posts/agent-eval-architecture/https://liliputtech.io/posts/agent-eval-architecture/Evaluation is an architecture layer: score the trajectory of model calls and tools, not the last sentence. Part 8 of How AI Agents Actually Work.Sun, 12 Jul 2026 00:00:00 GMThttps://liliputtech.io/posts/agent-orchestration/https://liliputtech.io/posts/agent-orchestration/Orchestration is the runtime loop: who calls the model, when tools run, and when to stop. Part 7 of How AI Agents Actually Work.Sat, 11 Jul 2026 00:00:00 GMThttps://liliputtech.io/posts/agent-rag/https://liliputtech.io/posts/agent-rag/RAG retrieves from an approved corpus, then grounds the next step. Part 6 of How AI Agents Actually Work; agentic RAG is retrieve-then-act.Fri, 10 Jul 2026 00:00:00 GMThttps://liliputtech.io/posts/agent-function-calling/https://liliputtech.io/posts/agent-function-calling/Function calling is the contract: the model emits a tool name and JSON arguments; the runtime executes. Part 5 of How AI Agents Actually Work.Thu, 09 Jul 2026 00:00:00 GMThttps://liliputtech.io/posts/agent-tools/https://liliputtech.io/posts/agent-tools/Tools are callable capabilities the runtime executes — APIs, search, code, files. Part 4 of How AI Agents Actually Work; the list of tools an agent can use, not the history.Wed, 08 Jul 2026 00:00:00 GMThttps://liliputtech.io/posts/agent-planning/https://liliputtech.io/posts/agent-planning/Planning turns a goal into steps and changes those steps after observations. Part 3 of How AI Agents Actually Work on re-planning inside the agent loop.Tue, 07 Jul 2026 00:00:00 GMThttps://liliputtech.io/posts/agent-memory/https://liliputtech.io/posts/agent-memory/Agent memory is what persists across steps: working context, session history, and longer-lived stores. Part 2 of How AI Agents Actually Work.Mon, 06 Jul 2026 00:00:00 GMThttps://liliputtech.io/posts/agent-llm/https://liliputtech.io/posts/agent-llm/The LLM is the language engine inside an agent — not the agent. Part 1 of How AI Agents Actually Work on next-token prediction, context, and what the model cannot do alone.Sun, 05 Jul 2026 00:00:00 GMThttps://liliputtech.io/posts/how-ai-agents-actually-work/https://liliputtech.io/posts/how-ai-agents-actually-work/An eight-part architecture map of AI agents: LLM, memory, planning, tools, function calling, RAG, orchestration, and evaluation — how the stack is wired.Sat, 04 Jul 2026 00:00:00 GMThttps://liliputtech.io/posts/agents-whats-next/https://liliputtech.io/posts/agents-whats-next/Near-term agents get more tools. Parts 1–9 mapped chat → reasoning → tools → loop → risk → fit → policy → eval → ship. This last part names patterns already peeking through. None of them require…Tue, 09 Jun 2026 00:00:00 GMThttps://liliputtech.io/posts/agent-shipping/https://liliputtech.io/posts/agent-shipping/Productising an agent is operations. Prompt engineering and policy said what it may be told and what it may touch. Evals said it behaves on the cases you care about. Shipping is how it lives:…Mon, 08 Jun 2026 00:00:00 GMThttps://liliputtech.io/posts/agent-evaluation/https://liliputtech.io/posts/agent-evaluation/Most production agents are not “trained.” They are evaluated until they behave. Part 7 wrote the specification. This part is about measuring the agent, again and again. Shipping waits for Part 9. If…Sun, 07 Jun 2026 00:00:00 GMThttps://liliputtech.io/posts/agent-policy/https://liliputtech.io/posts/agent-policy/Writing the agent is writing the prompt and the policy. Part 6 asked where agents fit. This part is design time: two documents that have to agree. The prompt is what the model is told (goal, scope,…Sat, 06 Jun 2026 00:00:00 GMThttps://liliputtech.io/posts/where-agents-make-sense/https://liliputtech.io/posts/where-agents-make-sense/Start where “done” is checkable and undo is possible. Part 5 was the cold water. This part is the filter. Agents are not a personality upgrade for every workflow. They pay off where a goal can be…Fri, 05 Jun 2026 00:00:00 GMThttps://liliputtech.io/posts/uncomfortable-agent-reality/https://liliputtech.io/posts/uncomfortable-agent-reality/Agents can still make mistakes — fluent, confident mistakes. The loop is powerful because it compounds. That is also why it is uncomfortable. A wrong token in chat is easy to ignore. A wrong token…Thu, 04 Jun 2026 00:00:00 GMThttps://liliputtech.io/posts/why-agents-are-different/https://liliputtech.io/posts/why-agents-are-different/Chat is a conversation. Agents are a control loop. Part 3 added tools. This part is why that is not “chat with plugins.” The way the system works changes. An error no longer stays in one paragraph.…Wed, 03 Jun 2026 00:00:00 GMThttps://liliputtech.io/posts/agent-era/https://liliputtech.io/posts/agent-era/Agents keep the language model, then add tools. Chat replies. Reasoning plans more carefully. An agent tries to finish a job: read a database, call an API, browse the web, run code, click through an…Tue, 02 Jun 2026 00:00:00 GMThttps://liliputtech.io/posts/reasoning-era/https://liliputtech.io/posts/reasoning-era/The next leap was not a friendlier chat box. It was more thinking time. In Part 1 the loop was prompt → fluent reply. That is fast, and it is often shallow on hard problems. The reasoning era is what…Mon, 01 Jun 2026 00:00:00 GMThttps://liliputtech.io/posts/chatbot-era/https://liliputtech.io/posts/chatbot-era/We spent years making AI better at answering questions. That habit still shapes how most people use the technology. You type. It replies. You edit. If it goes wrong, the damage is mostly words on a…Sun, 31 May 2026 00:00:00 GMThttps://liliputtech.io/posts/from-models-to-agents/https://liliputtech.io/posts/from-models-to-agents/A ten-part map from chatbots and reasoning models to tool-using AI agents — including prompt engineering, policy, evaluation, and shipping.Sat, 30 May 2026 00:00:00 GMThttps://liliputtech.io/posts/how-to-live-well-with-ai/https://liliputtech.io/posts/how-to-live-well-with-ai/You don’t need to become an engineer — you need a personal AI policy. Part 14 of AI Right Now on when to use AI, when to doubt it, and how to stay sharp.Fri, 15 May 2026 00:00:00 GMThttps://liliputtech.io/posts/what-comes-next/https://liliputtech.io/posts/what-comes-next/The near-term future is closer — and narrower — than sci-fi suggests. Part 13 of AI Right Now on agents, cheaper models, on-device AI, and overhyped forecasts.Thu, 14 May 2026 00:00:00 GMThttps://liliputtech.io/posts/ai-at-work/https://liliputtech.io/posts/ai-at-work/Companies bought AI; culture did not install itself. Part 12 of AI Right Now on pilots vs production, review, data rules, and workplace friction.Wed, 13 May 2026 00:00:00 GMThttps://liliputtech.io/posts/safety-trust-rules/https://liliputtech.io/posts/safety-trust-rules/Invention is outrunning trust. Part 11 of AI Right Now on safety layers: scams, privacy, over-confident advice, and the rules catching up.Tue, 12 May 2026 00:00:00 GMThttps://liliputtech.io/posts/money-power-ai-race/https://liliputtech.io/posts/money-power-ai-race/Behind every free AI chat is an expensive engine. Part 10 of AI Right Now on chips, data centers, distribution, and who actually pays for the race.Mon, 11 May 2026 00:00:00 GMThttps://liliputtech.io/posts/jobs-skills-division-of-labor/https://liliputtech.io/posts/jobs-skills-division-of-labor/AI is not coming for all jobs — it is coming for tasks inside jobs. Part 9 of AI Right Now on skills, role redesign, and the new division of labour.Sun, 10 May 2026 00:00:00 GMThttps://liliputtech.io/posts/where-ai-still-fails/https://liliputtech.io/posts/where-ai-still-fails/The most dangerous AI answer sounds perfect and is wrong. Part 8 of AI Right Now names failure modes: invented citations and false confidence.Sat, 09 May 2026 00:00:00 GMThttps://liliputtech.io/posts/where-ai-wins-today/https://liliputtech.io/posts/where-ai-wins-today/Ignore the hype and look at the results. Part 7 of AI Right Now names where AI already beats the old way — drafts, search, translation, code help.Fri, 08 May 2026 00:00:00 GMThttps://liliputtech.io/posts/ai-agents/https://liliputtech.io/posts/ai-agents/Chat answers; agents take steps. Part 6 of AI Right Now is about AI that executes work — and why control, approval, and audit become the product.Thu, 07 May 2026 00:00:00 GMThttps://liliputtech.io/posts/multimodal-ai/https://liliputtech.io/posts/multimodal-ai/AI left the chat box. Part 5 of AI Right Now covers vision, audio, and video — and why cameras and meetings raise new privacy stakes.Wed, 06 May 2026 00:00:00 GMThttps://liliputtech.io/posts/rise-of-large-language-models/https://liliputtech.io/posts/rise-of-large-language-models/Large language models made machines talk like us. Part 4 of AI Right Now explains fluency, invented citations, and why calm sentences are not proof.Tue, 05 May 2026 00:00:00 GMThttps://liliputtech.io/posts/how-modern-ai-learns/https://liliputtech.io/posts/how-modern-ai-learns/AI doesn’t know — it guesses from patterns. Part 3 of AI Right Now explains training, prediction, and why magical language gets us into trouble.Mon, 04 May 2026 00:00:00 GMThttps://liliputtech.io/posts/not-one-ai/https://liliputtech.io/posts/not-one-ai/Saying “AI” is like saying “vehicles.” Part 2 of AI Right Now splits the toolbox so you can tell generation from ranking, vision, and automation.Sun, 03 May 2026 00:00:00 GMThttps://liliputtech.io/posts/ai-right-now/https://liliputtech.io/posts/ai-right-now/AI is already in phones, maps, and inboxes. Part 1 of AI Right Now maps the quiet present — not the movie-robot future.Sat, 02 May 2026 00:00:00 GMThttps://liliputtech.io/posts/welcome-to-liliputtech/https://liliputtech.io/posts/welcome-to-liliputtech/Welcome to Liliputtech.io: a clear, unhyped map of AI today — what it can do, where it still fails, and how ordinary people can live with it wisely.Fri, 01 May 2026 00:00:00 GMT