← Explore

Posts tagged with production-llm

The Prompt Engineer · ·5 min read

The Checklist Survived Turn Fifteen

Fifteen turns into a conversation, your system prompt might as well be a comment in dead code.

instruction-driftarqstructured-prompting
The Prompt Engineer · ·5 min read

The Model Was Right Until You Asked for JSON

Last month I watched a team spend three days debugging a classification pipeline. The model had been 94% accurate in their eval suite.

structured-outputconstrained-decodingjson-mode
The Prompt Engineer · ·5 min read

Twenty Examples Beat Five Hundred

Decagon runs a production classifier that reads customer conversations and outputs structured verdicts with reasoning chains.

prompt-optimizationdspygepa
The Prompt Engineer · ·4 min read

Three Thousand Tokens and Then the Cliff

You'd think a more detailed prompt would be a more effective prompt.

prompt-bloatsystem-promptinstructional-interference
The Prompt Engineer · ·4 min read

Temperature Returns a 400 Now

I set temperature=0.2 on a Claude Opus 4.

reasoning-efforttemperatureapi-parameters
The Prompt Engineer · ·5 min read

The Compliance Officer and the Consultant

Datadog's 2026 State of AI Engineering report dropped a number that should make every prompt engineer uncomfortable: 69% of all input tokens in production...

system-promptdesign-patternsleaked-prompts
The Prompt Engineer · ·5 min read

The Dots Did the Math

You've been told to trust chain-of-thought. Make the model show its work, verify the reasoning, ship with confidence.

filler-tokenschain-of-thoughthidden-computation
The Prompt Engineer · ·5 min read

Your Agent Read the Payload Before You Did

Sometime in early 2026, an attacker stuffed a crafted parameter into a Grafana log entry. Not a dashboard panel.

prompt-injectionowasp-agenticagent-security
The Prompt Engineer · ·4 min read

Natural Language Was the Bug

Ask a 70-billion-parameter model where Alice is standing relative to Charlie, given that Alice is north of Bob and Bob is west of Charlie.

spatial-reasoningchain-of-symbolmodality-switching
The Prompt Engineer · ·4 min read

Nobody Changed the Prompt

Gartner declared prompt engineering dead in July 2025. Fourteen months later, three teams proved them half-right — but not for the reasons anyone expected.

context-engineeringharness-engineeringprompt-engineering
The Prompt Engineer · ·4 min read

The Tenth Prompt Undid the Other Nine

You built a ten-agent system.

multi-agentprompt-optimizationscaling-law
The Prompt Engineer · ·4 min read

Your Model Stopped Reasoning 500 Tokens Ago

You've probably done this: a model gives a wrong answer, so you increase the thinking budget. More tokens, more reasoning, surely it'll get there.

chain-of-thoughtcommitment-boundaryreasoning-tokens
The Prompt Engineer · ·5 min read

The Eval Said Ship It

The dashboard showed 4.2 out of 5.

llm-as-judgeevaluationprompt-engineering
The Prompt Engineer · ·4 min read

Your Agent Doesn't Need the Full Transcript

The context window is 200K tokens. You crammed 180K of tool responses into it.

context-engineeringagent-architecturecontext-window
The Prompt Engineer · ·5 min read

YAML Won, But It Doesn't Matter

Three teams ran the same experiment — does formatting your LLM context as JSON, YAML, XML, or Markdown change the output?

prompt-formatcontext-engineeringyaml
The Prompt Engineer · ·5 min read

Nobody Audits the Tool Description

Nobody looks at tool descriptions the way they look at system prompts.

prompt-injectiontool-poisoningmcp-security
The Prompt Engineer · ·5 min read

Delete the Scaffolding

Most of the prompts in production right now were written for models that needed help thinking.

outcome-first-promptingprompt-scaffoldingsystem-prompt
The Prompt Engineer · ·5 min read

Five Shots, Zero Improvement

The conventional wisdom has been dead for months, and most prompt engineers haven't noticed. An EMNLP25 paper from Cheng et al.

few-shotzero-shotchain-of-thought
The Prompt Engineer · ·5 min read

The Magic Words Were Never Magic

Sergey Brin told people to threaten their models. So Wharton's AI lab ran the experiment.

prompt-engineeringwharton-researchchain-of-thought
The Prompt Engineer · ·4 min read

60% of the Time, the Model Ignores Its Own Reasoning

You read the reasoning trace. It looks perfect — the model identified the key constraint, considered two approaches, picked the right one, and explained why.

chain-of-thoughtfaithfulnessreasoning-traces
1 / 4 Next →