Essays on AI product engineering, LLM evals, agents, prompts, and the napkin math of running models in production.https://mohithg.com/en-ushttps://mohithg.com/writing/ai-product-playbook-2026/https://mohithg.com/writing/ai-product-playbook-2026/If I were building a new AI product today, here's the order I'd do it in. Phase by phase, with the specific decisions that matter at each stage.Tue, 07 Jul 2026 00:00:00 GMTAI product engineeringhttps://mohithg.com/writing/ai-engineering-stack-2026/https://mohithg.com/writing/ai-engineering-stack-2026/AI product engineering has become a real discipline with its own stack. Here's the map I've ended up with after 18 months of writing about each piece, and where each piece fits.Mon, 06 Jul 2026 00:00:00 GMTAI product engineeringhttps://mohithg.com/writing/ai-deployment-rollouts/https://mohithg.com/writing/ai-deployment-rollouts/AI deployments have unique risks. Standard CI/CD patterns leave gaps. Here's the rollout discipline that catches problems before they reach all users.Sun, 05 Jul 2026 00:00:00 GMTAI infrastructurehttps://mohithg.com/writing/llm-load-testing/https://mohithg.com/writing/llm-load-testing/AI features fail differently under load than regular APIs. Standard load tests miss the failure modes that matter. Here's the load testing approach that finds real problems.Sat, 04 Jul 2026 00:00:00 GMTAI infrastructurehttps://mohithg.com/writing/multi-region-ai/https://mohithg.com/writing/multi-region-ai/Once your AI product has users worldwide, single-region deployment hurts. Multi-region adds complexity but solves real problems. Here's the architecture that works.Fri, 03 Jul 2026 00:00:00 GMTAI infrastructurehttps://mohithg.com/writing/llm-caching-layers/https://mohithg.com/writing/llm-caching-layers/Caching for LLM products has more layers than caching for regular APIs. Each layer has different tradeoffs. Here's the stack and the patterns that compound.Thu, 02 Jul 2026 00:00:00 GMTAI infrastructurehttps://mohithg.com/writing/streaming-llm-responses/https://mohithg.com/writing/streaming-llm-responses/Streaming the model's tokens to the user as they're generated dramatically improves perceived latency. The implementation has more gotchas than tutorials suggest.Wed, 01 Jul 2026 00:00:00 GMTAI infrastructurehttps://mohithg.com/writing/gpu-economics/https://mohithg.com/writing/gpu-economics/Self-hosting LLMs means renting GPUs. The cost calculation isn't just $/hour. Utilization, batching, quantization, and cold starts all change the picture. Here's the real math.Tue, 30 Jun 2026 00:00:00 GMTAI infrastructurehttps://mohithg.com/writing/inference-serving-frameworks/https://mohithg.com/writing/inference-serving-frameworks/If you're self-hosting LLMs, the inference server is one of the highest-leverage choices. Here's the landscape and the criteria that actually drive the decision.Mon, 29 Jun 2026 00:00:00 GMTAI infrastructurehttps://mohithg.com/writing/mcp-explained/https://mohithg.com/writing/mcp-explained/MCP is the protocol decoupling AI models from the tools and data they use. In 2026 it's becoming a baseline. Here's what it is and what to actually do about it.Sun, 28 Jun 2026 00:00:00 GMTAI infrastructurehttps://mohithg.com/writing/llm-gateway-pattern/https://mohithg.com/writing/llm-gateway-pattern/Calling LLM APIs directly from product code is fine until it isn't. The gateway pattern centralizes the cross-cutting concerns. Here's how to build one without overengineering.Sat, 27 Jun 2026 00:00:00 GMTAI infrastructurehttps://mohithg.com/writing/ai-infrastructure-decides-scale/https://mohithg.com/writing/ai-infrastructure-decides-scale/Prompts and models get attention. Infrastructure decides whether the product survives. Here's the infrastructure thinking that separates teams that scale from teams that don't.Fri, 26 Jun 2026 00:00:00 GMTAI infrastructurehttps://mohithg.com/writing/ai-abuse-detection/https://mohithg.com/writing/ai-abuse-detection/Some users will try to abuse your AI product. The volume of normal usage hides the abusive patterns until they're costly. Here's how to detect abuse without spying on legitimate users.Thu, 25 Jun 2026 00:00:00 GMTAI safety and guardrailshttps://mohithg.com/writing/ai-incident-response/https://mohithg.com/writing/ai-incident-response/AI incidents look different from regular software incidents. The playbook is similar but with AI-specific steps. Here's the runbook the teams I've seen use successfully.Wed, 24 Jun 2026 00:00:00 GMTAI safety and guardrailshttps://mohithg.com/writing/audit-trails-ai/https://mohithg.com/writing/audit-trails-ai/When something goes wrong with an AI system, the audit trail is what tells you what happened. Most AI systems don't have one. Here's the structure that holds up under investigation.Tue, 23 Jun 2026 00:00:00 GMTAI safety and guardrailshttps://mohithg.com/writing/refusal-design/https://mohithg.com/writing/refusal-design/Refusing user requests is part of every safe AI product. How the refusal is communicated determines whether users tolerate the limit or abandon the product. Here's the design.Mon, 22 Jun 2026 00:00:00 GMTAI safety and guardrailshttps://mohithg.com/writing/hallucination-mitigation/https://mohithg.com/writing/hallucination-mitigation/Eliminating hallucination is unrealistic. Preventing hallucinations from causing harm is achievable. Here's the reframing and the patterns that work.Sun, 21 Jun 2026 00:00:00 GMTAI safety and guardrailshttps://mohithg.com/writing/pii-handling-llm/https://mohithg.com/writing/pii-handling-llm/AI products handle user data. Most teams don't have a clear picture of where PII flows in their stack. Here's the audit and the patterns that actually keep data safe.Sat, 20 Jun 2026 00:00:00 GMTAI safety and guardrailshttps://mohithg.com/writing/jailbreak-resistance/https://mohithg.com/writing/jailbreak-resistance/Jailbreaks are attempts to make the AI ignore its constraints. They keep evolving. Defending against them requires more than the model's built-in resistance. Here's how.Fri, 19 Jun 2026 00:00:00 GMTAI safety and guardrailshttps://mohithg.com/writing/content-moderation-pipeline/https://mohithg.com/writing/content-moderation-pipeline/Models can produce content you don't want users to see. A moderation pipeline catches it before it reaches them. Here's the architecture and the patterns that work.Thu, 18 Jun 2026 00:00:00 GMTAI safety and guardrailshttps://mohithg.com/writing/red-teaming-ai/https://mohithg.com/writing/red-teaming-ai/The cheapest safety incident is the one you found yourself. Most teams don't red-team their AI products. Here's how to do it without a dedicated security team.Wed, 17 Jun 2026 00:00:00 GMTAI safety and guardrailshttps://mohithg.com/writing/prompt-injection-defense/https://mohithg.com/writing/prompt-injection-defense/Prompt injection gets discussed as a generic risk. The actual threats are specific and the defenses are specific. Here's the threat model and the defenses that work.Tue, 16 Jun 2026 00:00:00 GMTAI safety and guardrailshttps://mohithg.com/writing/ai-safety-as-engineering/https://mohithg.com/writing/ai-safety-as-engineering/Most AI safety conversations stay abstract. The teams shipping reliable AI products treat safety as concrete engineering: architecture, eval, instrumentation. Here's the discipline.Mon, 15 Jun 2026 00:00:00 GMTAI safety and guardrailshttps://mohithg.com/writing/freshness-in-rag/https://mohithg.com/writing/freshness-in-rag/A RAG system that returns yesterday's data on questions about today's reality is a liability. Keeping the index fresh is harder than it sounds. Here's the patterns.Sun, 14 Jun 2026 00:00:00 GMTRetrieval and RAGhttps://mohithg.com/writing/rag-with-permissions/https://mohithg.com/writing/rag-with-permissions/A multi-tenant RAG system has to enforce permissions at retrieval time, not after. Get this wrong and you have a data leak. Here's the architecture that holds up.Sat, 13 Jun 2026 00:00:00 GMTRetrieval and RAGhttps://mohithg.com/writing/long-context-vs-rag/https://mohithg.com/writing/long-context-vs-rag/Modern models support 200K+ token contexts. Some say RAG is dead. The reality is more nuanced. Here's the framing for when each approach actually wins.Fri, 12 Jun 2026 00:00:00 GMTRetrieval and RAGhttps://mohithg.com/writing/document-preprocessing-rag/https://mohithg.com/writing/document-preprocessing-rag/RAG systems are downstream of your document preprocessing. Bad text extraction, lost structure, broken tables: each one degrades retrieval. Here's the pipeline that matters.Thu, 11 Jun 2026 00:00:00 GMTRetrieval and RAGhttps://mohithg.com/writing/vector-db-choice/https://mohithg.com/writing/vector-db-choice/Vector DB choice gets discussed at length and decided poorly. Most teams pick by feature checklist; the actual tradeoffs are different. Here's the framework.Wed, 10 Jun 2026 00:00:00 GMTRetrieval and RAGhttps://mohithg.com/writing/query-rewriting-rag/https://mohithg.com/writing/query-rewriting-rag/User queries are not optimal retrieval queries. Rewriting the query before retrieval, often with an LLM, can dramatically improve recall. Most teams don't do it.Tue, 09 Jun 2026 00:00:00 GMTRetrieval and RAGhttps://mohithg.com/writing/rag-evaluation/https://mohithg.com/writing/rag-evaluation/Most teams evaluate the final answer their RAG system produces. That's necessary but not sufficient. Without evaluating retrieval separately, you can't tell what to fix.Mon, 08 Jun 2026 00:00:00 GMTRetrieval and RAGhttps://mohithg.com/writing/reranking-in-rag/https://mohithg.com/writing/reranking-in-rag/First-pass retrieval is fast and noisy. A reranker on top cleans up the order in tens of milliseconds. Skipping it leaves quality on the table.Sun, 07 Jun 2026 00:00:00 GMTRetrieval and RAGhttps://mohithg.com/writing/embedding-model-choice/https://mohithg.com/writing/embedding-model-choice/Your embedding model decision affects retrieval quality, cost, and the cost of every future migration. Most teams pick by leaderboard. Here's the decision that actually fits your product.Sat, 06 Jun 2026 00:00:00 GMTRetrieval and RAGhttps://mohithg.com/writing/hybrid-search/https://mohithg.com/writing/hybrid-search/Vector search is great until it isn't. The cases it misses are the ones BM25 catches. Combining both is the right default for most production RAG, and it's not as hard as it looks.Fri, 05 Jun 2026 00:00:00 GMTRetrieval and RAGhttps://mohithg.com/writing/chunking-strategies/https://mohithg.com/writing/chunking-strategies/How you split documents for retrieval is one of the highest-leverage RAG decisions and one of the most under-discussed. Here's the chunking playbook that actually works.Thu, 04 Jun 2026 00:00:00 GMTRetrieval and RAGhttps://mohithg.com/writing/retrieval-is-the-product/https://mohithg.com/writing/retrieval-is-the-product/Generative AI gets the attention. Retrieval does the work. The teams shipping reliable AI products spend most of their effort on the indexing, chunking, and ranking that nobody writes about.Wed, 03 Jun 2026 00:00:00 GMTRetrieval and RAGhttps://mohithg.com/writing/ai-product-quality-bar/https://mohithg.com/writing/ai-product-quality-bar/AI features are non-deterministic. They will make mistakes. The product question is how often, on which inputs, with what user-visible consequences. Here's the framework.Tue, 02 Jun 2026 00:00:00 GMTAI product engineeringhttps://mohithg.com/writing/ai-product-versioning/https://mohithg.com/writing/ai-product-versioning/AI product behavior changes when models change. Users notice. The versioning model determines who absorbs the change. Get this wrong and your users feel like the product is randomly different.Mon, 01 Jun 2026 00:00:00 GMTAI product engineeringhttps://mohithg.com/writing/from-demo-to-production/https://mohithg.com/writing/from-demo-to-production/A working AI demo is maybe 20% of the work. The other 80% is everything that makes it survive contact with real users. Here's the punch list.Sun, 31 May 2026 00:00:00 GMTAI product engineeringhttps://mohithg.com/writing/ai-product-team-shape/https://mohithg.com/writing/ai-product-team-shape/Building AI products requires combinations of skill that don't fit traditional team structures. Here's the team shape that actually works and the dysfunction patterns to avoid.Sat, 30 May 2026 00:00:00 GMTAI product engineeringhttps://mohithg.com/writing/ai-product-roadmap/https://mohithg.com/writing/ai-product-roadmap/Traditional roadmaps assume the technology underneath is stable. AI products live on a substrate that changes every few months. Here's the planning approach that adapts.Fri, 29 May 2026 00:00:00 GMTAI product engineeringhttps://mohithg.com/writing/ai-product-trust/https://mohithg.com/writing/ai-product-trust/AI features have a trust problem most software features don't. Users have learned to be skeptical. The features that earn trust do specific things. Here's the list.Thu, 28 May 2026 00:00:00 GMTAI product engineeringhttps://mohithg.com/writing/measuring-ai-product-success/https://mohithg.com/writing/measuring-ai-product-success/Most AI product dashboards track the wrong things. Engagement is misleading; AI-feature usage is decoration. Here are the metrics that actually tell you whether your AI feature is working.Wed, 27 May 2026 00:00:00 GMTAI product engineeringhttps://mohithg.com/writing/ai-feature-flags/https://mohithg.com/writing/ai-feature-flags/AI features fail differently from regular features. Standard rollout patterns leave you exposed to model regressions and traffic-driven failures. Here's the gating model that fits.Tue, 26 May 2026 00:00:00 GMTAI product engineeringhttps://mohithg.com/writing/ai-product-onboarding/https://mohithg.com/writing/ai-product-onboarding/First-touch experience determines whether users come back. AI products have a unique onboarding problem: managing expectations the model may or may not meet. Here's the playbook.Mon, 25 May 2026 00:00:00 GMTAI product engineeringhttps://mohithg.com/writing/ai-features-that-disappear/https://mohithg.com/writing/ai-features-that-disappear/The best AI features in 2026 don't have an 'AI' label. They're invisible improvements to existing flows. Here's why most AI-branded features fail and the disappearing ones succeed.Sun, 24 May 2026 00:00:00 GMTAI product engineeringhttps://mohithg.com/writing/optimizing-llm-spend-late-stage/https://mohithg.com/writing/optimizing-llm-spend-late-stage/Most cost-optimization advice assumes you're starting from scratch. What if you already have a $100K/month bill and need to bring it down without breaking the product? Here's the order of operations.Sat, 23 May 2026 00:00:00 GMTThe napkin math of AI in productionhttps://mohithg.com/writing/llm-batch-vs-realtime/https://mohithg.com/writing/llm-batch-vs-realtime/Many LLM workloads that run synchronously in production should be running asynchronously, and vice versa. The cost and reliability difference is large. Here's the framing.Fri, 22 May 2026 00:00:00 GMTThe napkin math of AI in productionhttps://mohithg.com/writing/cost-attribution-llm/https://mohithg.com/writing/cost-attribution-llm/An aggregate API bill tells you nothing about which features, users, or queries drive cost. Without attribution, you can't optimize. Here's the model that works.Thu, 21 May 2026 00:00:00 GMTThe napkin math of AI in productionhttps://mohithg.com/writing/llm-build-vs-buy/https://mohithg.com/writing/llm-build-vs-buy/Should you build your own model, fine-tune, host open-source, or call APIs? The decision depends on a few specific questions, and the answer is usually 'call APIs.'Wed, 20 May 2026 00:00:00 GMTThe napkin math of AI in productionhttps://mohithg.com/writing/llm-rate-limits/https://mohithg.com/writing/llm-rate-limits/Provider rate limits constrain what you can ship more often than they should. Most teams hit the limits at the wrong time and don't have a plan. Here's the planning framework.Tue, 19 May 2026 00:00:00 GMTThe napkin math of AI in productionhttps://mohithg.com/writing/cost-of-context/https://mohithg.com/writing/cost-of-context/Long context windows let you stuff more into a prompt. They don't let you do it for free. The cost scales superlinearly with context size in ways that surprise teams.Mon, 18 May 2026 00:00:00 GMTThe napkin math of AI in productionhttps://mohithg.com/writing/llm-pricing-tier-design/https://mohithg.com/writing/llm-pricing-tier-design/Flat-rate AI pricing leaves you exposed to the heavy users. Pure pay-per-use is hostile to most users. The middle ground is tiers with clear limits, designed around your cost distribution.Sun, 17 May 2026 00:00:00 GMTThe napkin math of AI in productionhttps://mohithg.com/writing/model-routing-cost/https://mohithg.com/writing/model-routing-cost/Not every request needs the frontier model. Routing requests to the right model tier is one of the highest-leverage cost optimizations and one of the most underused.Sat, 16 May 2026 00:00:00 GMTThe napkin math of AI in productionhttps://mohithg.com/writing/prompt-caching-economics/https://mohithg.com/writing/prompt-caching-economics/Modern LLM APIs let you cache the static parts of your prompt. Most teams enable it, then design prompts that defeat it. Here's how to get the actual savings.Fri, 15 May 2026 00:00:00 GMTThe napkin math of AI in productionhttps://mohithg.com/writing/llm-unit-economics/https://mohithg.com/writing/llm-unit-economics/Unit economics for LLM features look different from regular software unit economics. The variable costs are real, the gross margins can flip with usage patterns, and the questions are coming. Here's how to think about them.Thu, 14 May 2026 00:00:00 GMTThe napkin math of AI in productionhttps://mohithg.com/writing/agent-cost-control/https://mohithg.com/writing/agent-cost-control/An agent that costs $0.10 per run becomes a $30K monthly bill at meaningful traffic. Here's where the cost concentrates and which controls keep it sustainable.Wed, 13 May 2026 00:00:00 GMTAgent architecturehttps://mohithg.com/writing/agent-observability/https://mohithg.com/writing/agent-observability/An agent without observability is a black box that occasionally produces output. Here's what to instrument, what to alert on, and what to keep out of your dashboards.Tue, 12 May 2026 00:00:00 GMTAgent architecturehttps://mohithg.com/writing/agent-latency/https://mohithg.com/writing/agent-latency/An agent that takes 30 seconds to answer is unusable for most product surfaces. Here's where the time actually goes and which optimizations move the needle.Mon, 11 May 2026 00:00:00 GMTAgent architecturehttps://mohithg.com/writing/agent-tool-permissions/https://mohithg.com/writing/agent-tool-permissions/An agent with the wrong tool permissions is a security incident waiting to happen. Here's the permission model that keeps agents capable without giving them the keys to everything.Sun, 10 May 2026 00:00:00 GMTAgent architecturehttps://mohithg.com/writing/agent-evals/https://mohithg.com/writing/agent-evals/Eval frameworks for single-prompt LLM features don't translate cleanly to agents. Agents have process. The bench needs to grade the process, not just the result.Sat, 09 May 2026 00:00:00 GMTAgent architecturehttps://mohithg.com/writing/multi-agent-vs-single-agent/https://mohithg.com/writing/multi-agent-vs-single-agent/Multi-agent architectures look elegant in diagrams. In production, they're more often a tax than a benefit. Here's when the orchestra actually beats the soloist.Fri, 08 May 2026 00:00:00 GMTAgent architecturehttps://mohithg.com/writing/agent-failure-modes/https://mohithg.com/writing/agent-failure-modes/Production agents fail in predictable ways. Knowing the patterns saves weeks of debugging. Here are the five I see most often and what actually fixes them.Thu, 07 May 2026 00:00:00 GMTAgent architecturehttps://mohithg.com/writing/when-to-use-agents/https://mohithg.com/writing/when-to-use-agents/The 'agent' label has been applied to almost every LLM feature. Most of them shouldn't be agents. Here's the actual decision criteria.Wed, 06 May 2026 00:00:00 GMTAgent architecturehttps://mohithg.com/writing/agent-state-management/https://mohithg.com/writing/agent-state-management/Most agent tutorials skip past the question of where state lives. In production, state management is half the work. Here's the model that scales.Tue, 05 May 2026 00:00:00 GMTAgent architecturehttps://mohithg.com/writing/tool-design-for-agents/https://mohithg.com/writing/tool-design-for-agents/An agent is only as good as the tools you give it. Most teams design tools the way they design APIs for other engineers, and pay for it. Here's the difference that matters.Mon, 04 May 2026 00:00:00 GMTAgent architecturehttps://mohithg.com/writing/the-ai-vocabulary-problem/https://mohithg.com/writing/the-ai-vocabulary-problem/Every word your LLM is allowed to say imposes obligations on the systems beneath it. Treat the prompt's vocabulary like an interface or pay for it later.Sun, 03 May 2026 00:00:00 GMTPrompts as API contractshttps://mohithg.com/writing/human-in-the-loop-evals/https://mohithg.com/writing/human-in-the-loop-evals/Automated evals can do a lot, but not everything. Here's where humans still beat any LLM judge, and how to set up the human review loop without breaking the bank.Sun, 03 May 2026 00:00:00 GMTLLM eval engineeringhttps://mohithg.com/writing/llm-eval-bench-actually-needs/https://mohithg.com/writing/llm-eval-bench-actually-needs/Most eval frameworks measure whether the model returned a string. Production eval benches measure whether shipping the change is safe. The gap is everything.Sat, 02 May 2026 00:00:00 GMTLLM eval engineeringhttps://mohithg.com/writing/adversarial-evals/https://mohithg.com/writing/adversarial-evals/The friendly cases will tell you the model usually works. The adversarial cases will tell you what happens when things go wrong. Most teams don't have enough of the second kind.Sat, 02 May 2026 00:00:00 GMTLLM eval engineeringhttps://mohithg.com/writing/eval-drift/https://mohithg.com/writing/eval-drift/An eval bench can pass with flying colors while production quality declines. The gap is called eval drift, and it's the most common silent failure in LLM ops.Fri, 01 May 2026 00:00:00 GMTLLM eval engineeringhttps://mohithg.com/writing/agent-loops-are-graphs/https://mohithg.com/writing/agent-loops-are-graphs/Strip away the agent terminology and you're left with a graph of function calls with conditional edges. The patterns that ship treat them that way.Fri, 01 May 2026 00:00:00 GMTAgent architecturehttps://mohithg.com/writing/hidden-cost-of-evals/https://mohithg.com/writing/hidden-cost-of-evals/Eval pipelines are easy to start and expensive to run at scale. Here's where the cost actually comes from and how to keep it under control without losing the safety net.Thu, 30 Apr 2026 00:00:00 GMTLLM eval engineeringhttps://mohithg.com/writing/economics-of-llm-in-production/https://mohithg.com/writing/economics-of-llm-in-production/Napkin math for the unit cost of an AI feature: tokens, latency, caching, model routing, and the surprising line items nobody publishes.Thu, 30 Apr 2026 00:00:00 GMTThe napkin math of AI in productionhttps://mohithg.com/writing/eval-datasets-that-hold-up/https://mohithg.com/writing/eval-datasets-that-hold-up/Most eval datasets rot. The cases drift, the rubrics get stale, the bench becomes a museum piece. Here's how to build one that stays useful for years.Wed, 29 Apr 2026 00:00:00 GMTLLM eval engineeringhttps://mohithg.com/writing/build-substance-then-surface/https://mohithg.com/writing/build-substance-then-surface/Most AI product failures are LLM wrappers shipped before there's anything underneath worth wrapping. The hard part of an AI product is almost never the prompt.Wed, 29 Apr 2026 00:00:00 GMTAI product engineeringhttps://mohithg.com/writing/prompts-as-type-signatures/https://mohithg.com/writing/prompts-as-type-signatures/The quickest mental model improvement for prompt engineering: stop thinking of prompts as instructions, start thinking of them as type signatures for the model's output.Tue, 28 Apr 2026 00:00:00 GMTPrompts as API contractshttps://mohithg.com/writing/continuous-vs-deep-vs-shadow-evals/https://mohithg.com/writing/continuous-vs-deep-vs-shadow-evals/Most teams treat 'evals' as one thing. The teams shipping reliable AI products run three distinct eval loops at different cadences. Here's the breakdown.Tue, 28 Apr 2026 00:00:00 GMTLLM eval engineeringhttps://mohithg.com/writing/system-prompts-that-age-well/https://mohithg.com/writing/system-prompts-that-age-well/A system prompt is shipped code. It needs the same discipline. Here are the patterns that survive a year of model upgrades, prompt edits, and team turnover.Mon, 27 Apr 2026 00:00:00 GMTPrompts as API contractshttps://mohithg.com/writing/eval-rubrics-are-the-work/https://mohithg.com/writing/eval-rubrics-are-the-work/Most teams treat the eval rubric as paperwork. The teams shipping reliable LLM products treat the rubric as the actual product specification. Here's the difference.Mon, 27 Apr 2026 00:00:00 GMTLLM eval engineeringhttps://mohithg.com/writing/prompt-versioning-practical/https://mohithg.com/writing/prompt-versioning-practical/Versioning prompts is harder than versioning code because the artifact is a string and the test suite is fuzzy. Here's the workflow that ships.Sun, 26 Apr 2026 00:00:00 GMTPrompts as API contractshttps://mohithg.com/writing/llm-as-judge-what-works/https://mohithg.com/writing/llm-as-judge-what-works/Using one LLM to grade another LLM's output is the most over-deployed and under-evaluated eval pattern in production. Here's when it works, when it fails, and how to use it well.Sun, 26 Apr 2026 00:00:00 GMTLLM eval engineeringhttps://mohithg.com/writing/when-to-ship-a-prompt-change/https://mohithg.com/writing/when-to-ship-a-prompt-change/The decision rule that separates teams who ship prompt changes confidently from teams who hover their finger over the button.Sat, 25 Apr 2026 00:00:00 GMTPrompts as API contractshttps://mohithg.com/writing/minimum-viable-eval-bench/https://mohithg.com/writing/minimum-viable-eval-bench/Most LLM teams ship without a real eval bench. The reason isn't that benches are hard. It's that the first one feels too small to matter. Here's the smallest useful one.Sat, 25 Apr 2026 00:00:00 GMTLLM eval engineeringhttps://mohithg.com/writing/few-shot-design/https://mohithg.com/writing/few-shot-design/Few-shot examples are the most reliable way to shape model behavior. Most production prompts use them badly or skip them entirely. Here's how to use them well.Fri, 24 Apr 2026 00:00:00 GMTPrompts as API contractshttps://mohithg.com/writing/schema-first-prompts/https://mohithg.com/writing/schema-first-prompts/Constrained generation, structured output APIs, and JSON schema have made prompt engineering more like API design and less like creative writing. Lean in.Thu, 23 Apr 2026 00:00:00 GMTPrompts as API contractshttps://mohithg.com/writing/personas-useful-or-theatre/https://mohithg.com/writing/personas-useful-or-theatre/Almost every system prompt starts with 'You are a helpful assistant.' Most personas in prompts are decorative. Here's when they actually move the needle, and when they're padding.Wed, 22 Apr 2026 00:00:00 GMTPrompts as API contractshttps://mohithg.com/writing/trace-everything-debugging/https://mohithg.com/writing/trace-everything-debugging/You cannot debug what you cannot replay. The single highest-leverage habit in LLM engineering is making every model call inspectable after the fact.Tue, 21 Apr 2026 00:00:00 GMTPrompts as API contractshttps://mohithg.com/writing/system-vs-user-vs-developer/https://mohithg.com/writing/system-vs-user-vs-developer/Modern LLM APIs distinguish between system, user, developer, and assistant roles. The rules for which content goes in which slot aren't intuitive. Here's the working model.Sun, 19 Apr 2026 00:00:00 GMTPrompts as API contracts