{"version":"https://jsonfeed.org/version/1.1","title":"OpenFactoryAI blog","home_page_url":"https://blog.openfactoryai.com","feed_url":"https://blog.openfactoryai.com/feed.json","items":[{"id":"https://blog.openfactoryai.com/the-last-seat","url":"https://blog.openfactoryai.com/the-last-seat","title":"The Last Seat","summary":"The rise and fall of software GCCs and ODCs is not the disappearance of global engineering. It is the end of the seat, person-month, and billable pyramid as the default unit of software capacity.","date_published":"2026-07-14T21:00:00.289665+00:00","tags":["Global Capability Centers","Offshore Development Centers","software factory AI","offshoring economics","CTO strategy"]},{"id":"https://blog.openfactoryai.com/swarms-multiply-errors","url":"https://blog.openfactoryai.com/swarms-multiply-errors","title":"Swarms Multiply Errors","summary":"Multi-agent systems create useful search diversity only when roles, evidence, state ownership, communication, merge authority, cancellation, and verification are engineered explicitly.","date_published":"2026-07-14T20:59:00.289665+00:00","tags":["multi-agent systems","AI swarms","agent orchestration","parallel agents","software automation"]},{"id":"https://blog.openfactoryai.com/the-last-step-gets-too-much-credit","url":"https://blog.openfactoryai.com/the-last-step-gets-too-much-credit","title":"The Last Step Gets Too Much Credit","summary":"Long agent workflows need temporal, structural, and multi-agent credit assignment that distinguishes decisive work from visible final actions without rewarding plausible narration.","date_published":"2026-07-14T20:58:00.289665+00:00","tags":["credit assignment","reinforcement learning agents","multi-agent systems","process rewards","agent workflows"]},{"id":"https://blog.openfactoryai.com/reward-is-a-bug-report","url":"https://blog.openfactoryai.com/reward-is-a-bug-report","title":"Reward Is a Bug Report","summary":"Reinforcement learning for agents succeeds or fails on environment state, action contracts, reset fidelity, verifiers, reward timing, and the shortcuts the reward accidentally permits.","date_published":"2026-07-14T20:57:00.289665+00:00","tags":["reinforcement learning agents","agent training","reward design","multi-hop workflows","AI environments"]},{"id":"https://blog.openfactoryai.com/the-benchmark-ends-too-soon","url":"https://blog.openfactoryai.com/the-benchmark-ends-too-soon","title":"The Benchmark Ends Too Soon","summary":"Agent evaluation must measure the production task after generation: execution, verification, rework, review, latency, cost, recovery, and value under the real workload.","date_published":"2026-07-14T20:56:00.289665+00:00","tags":["AI agent evaluation","SWE-bench","LLM benchmarks","production evaluation","software automation ROI"]},{"id":"https://blog.openfactoryai.com/no-trace-no-truth","url":"https://blog.openfactoryai.com/no-trace-no-truth","title":"No Trace. No Truth.","summary":"Agent observability must connect prompts, tokens, retrieval, tools, state transitions, costs, policy decisions, artifacts, and verification to one terminal outcome.","date_published":"2026-07-14T20:55:00.289665+00:00","tags":["AI agent observability","distributed tracing","OpenTelemetry","inference monitoring","Loop Engineering"]},{"id":"https://blog.openfactoryai.com/after-the-demo","url":"https://blog.openfactoryai.com/after-the-demo","title":"After the Demo","summary":"AI-native development begins when an agent task must survive timeouts, retries, process death, duplicate delivery, changing repositories, and partial side effects without losing the truth.","date_published":"2026-07-14T20:54:00.289665+00:00","tags":["AI-native development","durable execution","agent workflows","idempotency","software agents"]},{"id":"https://blog.openfactoryai.com/tokenmaxing","url":"https://blog.openfactoryai.com/tokenmaxing","title":"TokenMaxing","summary":"TokenMaxing is not asking every model to think longer. It allocates retries, branches, critiques, context, and verification where the next unit of inference has positive marginal verified-outcome value.","date_published":"2026-07-14T20:53:00.289665+00:00","tags":["TokenMaxing","test-time compute","agent retries","inference economics","Loop Engineering"]},{"id":"https://blog.openfactoryai.com/the-loop-is-the-runtime","url":"https://blog.openfactoryai.com/the-loop-is-the-runtime","title":"The Loop Is the Runtime","summary":"An agent is a stateful runtime around inference. Define observe, decide, act, verify, stop, authority, budgets, event logs, recovery, and a complete eight-call workflow trace before tuning the prompt.","date_published":"2026-07-14T20:52:00.289665+00:00","tags":["Loop Engineering","AI agents","agent runtime","tool use","state machines"]},{"id":"https://blog.openfactoryai.com/make-invalid-tokens-impossible","url":"https://blog.openfactoryai.com/make-invalid-tokens-impossible","title":"Make Invalid Tokens Impossible","summary":"Constrained decoding masks tokens that cannot lead to valid JSON, SQL, or tool calls. Learn the syntax/semantics boundary, tokenizer alignment, retry economics, schema design, validation, authorization, and safe execution.","date_published":"2026-07-14T20:51:00.289665+00:00","tags":["constrained decoding","structured output","JSON schema","tool calling","agent safety"]},{"id":"https://blog.openfactoryai.com/rag-has-seven-places-to-lie","url":"https://blog.openfactoryai.com/rag-has-seven-places-to-lie","title":"Seven Places RAG Can Lie","summary":"RAG is a control system across ingestion, retrieval, reranking, context, generation, citations, and feedback. Trace each failure, measure conditional stage quality, and stop blaming the vector database for every wrong answer.","date_published":"2026-07-14T20:50:00.289665+00:00","tags":["RAG","retrieval augmented generation","vector search","groundedness","citations"]},{"id":"https://blog.openfactoryai.com/context-is-a-budget","url":"https://blog.openfactoryai.com/context-is-a-budget","title":"The Window Is Not the Memory","summary":"A large context window is capacity, not a mandate to fill it. Allocate tokens across policy, evidence, tools, history, and output; test position, distractors, compression, latency, and verified outcome quality.","date_published":"2026-07-14T20:49:00.289665+00:00","tags":["context window","long context","prompt compression","agents","retrieval"]},{"id":"https://blog.openfactoryai.com/route-the-work","url":"https://blog.openfactoryai.com/route-the-work","title":"Route the Work","summary":"An LLM router is a policy under uncertainty. Compare fixed models, cascades, and learned routing using expected outcome cost, calibration, privacy, latency, availability, and a reproducible 1,000-request scenario.","date_published":"2026-07-14T20:48:00.289665+00:00","tags":["LLM routing","model cascade","RouteLLM","FrugalGPT","cost optimization"]},{"id":"https://blog.openfactoryai.com/quantization-without-magic","url":"https://blog.openfactoryai.com/quantization-without-magic","title":"Four Bits. Full Consequences.","summary":"Quantization changes the precision of weights, activations, or KV state. Calculate the real memory budget, understand kernels and calibration, and validate task-specific quality before claiming faster inference.","date_published":"2026-07-14T20:47:00.289665+00:00","tags":["LLM quantization","GPTQ","AWQ","SmoothQuant","KV cache"]},{"id":"https://blog.openfactoryai.com/speculative-decoding","url":"https://blog.openfactoryai.com/speculative-decoding","title":"The Draft Is Allowed to Be Wrong","summary":"Speculative decoding uses cheap draft work and parallel target verification to advance several output tokens per target step. Derive the breakeven, preserve the target distribution, and test acceptance, memory, latency, and throughput.","date_published":"2026-07-14T20:46:00.289665+00:00","tags":["speculative decoding","LLM inference","draft model","decoding latency","inference optimization"]},{"id":"https://blog.openfactoryai.com/continuous-batching","url":"https://blog.openfactoryai.com/continuous-batching","title":"The Batch Never Waits","summary":"Continuous batching schedules LLM inference one token iteration at a time. See how requests enter and leave a live batch, why prefills stall decodes, and how fairness and memory admission become product policy.","date_published":"2026-07-14T20:45:00.289665+00:00","tags":["continuous batching","LLM serving","GPU scheduling","inference throughput","fairness"]},{"id":"https://blog.openfactoryai.com/the-cache-ladder","url":"https://blog.openfactoryai.com/the-cache-ladder","title":"The Cache Ladder","summary":"LLM caching is four different architectural decisions. Learn what exact, semantic, prefix, and KV caches reuse, how to key and invalidate them, and when a hit costs more than a miss.","date_published":"2026-07-14T20:44:00.289665+00:00","tags":["LLM caching","semantic cache","prefix cache","KV cache","inference optimization"]},{"id":"https://blog.openfactoryai.com/capacity-is-a-queue","url":"https://blog.openfactoryai.com/capacity-is-a-queue","title":"Capacity Is a Queue","summary":"Fast models can serve slow products when bursts, long prompts, agent fan-out, batching, memory pressure, and high utilization create queues. Use queueing math, goodput, and admission control to plan inference capacity.","date_published":"2026-07-14T20:43:00.289665+00:00","tags":["inference serving","queueing","capacity planning","tail latency","autoscaling"]},{"id":"https://blog.openfactoryai.com/cost-per-verified-outcome","url":"https://blog.openfactoryai.com/cost-per-verified-outcome","title":"The Token Bill Lies","summary":"Token prices are an input cost, not an automation business case. Build the full unit economics across model calls, tools, retries, review, recovery, false acceptance, and fixed integration cost.","date_published":"2026-07-14T20:42:00.289665+00:00","tags":["inference economics","automation roi","cost per outcome","routing","agent reliability"]},{"id":"https://blog.openfactoryai.com/from-prompt-to-proof","url":"https://blog.openfactoryai.com/from-prompt-to-proof","title":"From Prompt to Proof","summary":"Follow one automated request through gateway, routing, context, queueing, prefill, decoding, tools, verification, and commit. The important latency is not first token. It is time to a verified outcome.","date_published":"2026-07-14T20:41:00.289665+00:00","tags":["inference","agent runtime","observability","latency","verification"]},{"id":"https://blog.openfactoryai.com/three-kinds-ai-factory","url":"https://blog.openfactoryai.com/three-kinds-ai-factory","title":"Three Factories. One Missing Layer.","summary":"AI factory can mean a GPU plant, a model-production system, or a software-delivery system. Here is the map, the current software-factory landscape, and why OpenFactoryAI sits outside the line as proof.","date_published":"2026-07-14T20:40:00.289665+00:00","tags":["ai factory","software factory","coding agents","verification","india software services"]},{"id":"https://blog.openfactoryai.com/who-signs-off-on-ai-written-code","url":"https://blog.openfactoryai.com/who-signs-off-on-ai-written-code","title":"Who signs off on AI written code?","summary":"When an agent writes the change, a human still has to vouch for it. Today that human never gets to go home. Here is what replaces them.","date_published":"2026-07-04T00:00:00+00:00","tags":["sign off","accountability","audit","agents","certification"]},{"id":"https://blog.openfactoryai.com/what-a-factory-must-prove","url":"https://blog.openfactoryai.com/what-a-factory-must-prove","title":"What a factory must prove","summary":"The five things any agent built artifact has to carry before you trust it, and why each one ships as evidence rather than as a promise.","date_published":"2026-07-04T00:00:00+00:00","tags":["trust pillars","provenance","audit","maintainability","verification"]},{"id":"https://blog.openfactoryai.com/code-got-free-trust-didnt","url":"https://blog.openfactoryai.com/code-got-free-trust-didnt","title":"Code got free. Trust didn't.","summary":"Models generate fluent software cheaply, so the bottleneck moved. The unsolved problem is proving the output is safe to ship.","date_published":"2026-07-04T00:00:00+00:00","tags":["verification","agents","trust","software factories"]}]}