#ai-engineering

Articles tagged with #ai-engineering

Every Screenshot You Paste Is Billed Again on Every Turn
Pasting a screenshot into an AI chat feels free. You drop it in, ask a question, get an answer. But an image is not a file attachment in the way an email attachment is. It is input tokens, and in a lo
Oct 13, 202611 min read1
The Hidden Cost of a Long AI Conversation (and 5 Ways to Cut It)
You ask a follow-up question and the answer comes back in seconds. It feels like one more small message. For the model, it is the whole conversation again, from the first line to your latest question.
Oct 12, 20268 min read4
Your LLM App Will Hit a 429. Here's Why Retrying Is the Wrong Fix
Every LLM app hits a 429 in production. Most retry code treats all of them the same: wait, try again, hope. The docs say that is wrong in at least three ways, and streaming adds a fourth. How to read
Oct 12, 20268 min read1
Your MCP Servers Bill You on Every Request: The Tool-Definition Tax
You connect a few MCP servers and your agent gains dozens of abilities. What the connection also adds is a block of text that rides along in the prompt of every single request, whether the agent uses
Oct 10, 20267 min read4
Reasoning Effort Across AI Providers: One Knob, Four Meanings
How effort and thinking controls differ at Anthropic, OpenAI, DeepSeek and Qwen, what they cost, and why changing effort breaks your cache.
Oct 9, 20268 min read7
How AI Chat "Compaction" Works: Summaries That Shrink the Context
How compaction swaps old turns for a summary, what it loses, and when it saves money, from Anthropic's docs.
Oct 9, 20269 min read1