Bye Prompt
After the prompt box
Chat vs workflow vs agent — pick a mode, copy a spec, and stop unsupervised spend. A compact field guide to operating AI after the blank prompt box. Tabaconda LLC, Florida.
Featured guides
Editor’s picks — start here, then browse by topic.
Try it: pick a mode, copy a spec
Chat, workflow, or agent. The output is a starter SPEC.md — not a personality prompt.
Three modes
The three ways to work with AI in 2026: a chat, a saved workflow, and an agent with tools. Pick the mode on purpose or you will fight the tool.
Read
Context packs
A context pack is the files, constraints, examples, and non-goals you assemble before an agent or model runs. It beats a clever paragraph.
Read
Compiled specs
A super prompt is not a longer paragraph. It is a spec compiled from your brief, files, examples, and non-goals. How to build one without folklore.
ReadAlso worth your time
Loops
Agent work is a loop: plan, act, critique against the spec, retry or stop. How to run that loop without unsupervised intern energy.
Evaluation
If you cannot test done, you do not have an agent workflow. Acceptance tests, traces, retry budgets, and what ‘looks good’ is hiding.
Agents in 2026
A dated field guide to AI agents in 2026: computer-use, operators, assistants that click, and how to read vendor claims without swallowing the keynote.
Tool scorecard
A reusable rubric for AI agent and assistant products: actions, traces, caps, eval, data, and lock-in. Not a paid ranking.
Checklists
Printable-in-spirit checklists: pick a mode, build a pack, run a loop, promote an agent. Steal and modify.
Editorial
Who publishes Bye Prompt, how we research agent workflows, and how to send corrections — Tabaconda LLC, Florida, USA.
Start Here
3 guides Three modes The three ways to work with AI in 2026: a chat, a saved workflow, and an agent with tools. Pick the mode on purpose or you will fight the tool. Open Catalog Bye Prompt sits at the center of a small catalog: Create, Pay, Buy, Car, Homes. Not a forty-site footer. Open Checklists Printable-in-spirit checklists: pick a mode, build a pack, run a loop, promote an agent. Steal and modify. Open Operating
7 guides Context packs A context pack is the files, constraints, examples, and non-goals you assemble before an agent or model runs. It beats a clever paragraph. Open Compiled specs A super prompt is not a longer paragraph. It is a spec compiled from your brief, files, examples, and non-goals. How to build one without folklore. Open Loops Agent work is a loop: plan, act, critique against the spec, retry or stop. How to run that loop without unsupervised intern energy. Open Evaluation If you cannot test done, you do not have an agent workflow. Acceptance tests, traces, retry budgets, and what ‘looks good’ is hiding. Open Human in the loop Agents should not own irreversible steps. Where to put a human: money, publish, production, identity, and taste. Open What not to automate A practical deny-list for agents: irreversible money, identity, production, private data, and jobs you cannot evaluate. Open Memory & tools Models do not remember your company. Files, tools, and traces are memory. How to give an agent a world without dumping your disk. Open Landscape
4 guides Agents in 2026 A dated field guide to AI agents in 2026: computer-use, operators, assistants that click, and how to read vendor claims without swallowing the keynote. Open Ambient vs agent Ambient assistants suggest; agents act. Mixing the two words is how products over-promise. When ‘it just knows’ is a lie. Open When you still type Prompting is not dead. It is a subroutine: repairs, tiny jobs, compile steps, and anything cheaper than a workflow. How to type on purpose. Open Tool scorecard A reusable rubric for AI agent and assistant products: actions, traces, caps, eval, data, and lock-in. Not a paid ranking. Open Patterns
7 guides Research A workflow for AI-assisted research: pack the sources, compile a spec, fetch, table, cite, stop. No invented bibliography. Open Making How Bye Prompt thinks about creative production with AI: briefs, references, loops, finish. Deep studio how-to lives on Create By Prompt. Open Buying How to use AI for purchase decisions without unsupervised checkout: spec, matrix, buy/wait/skip, human confirm. Depth on Buy By Prompt. Open Money Agents and money: drafts, caps, identity, traces. Bye Prompt will not tell you to wire blindly. Pay By Prompt is the field guide for rails. Open Ops Using agents in operations: runbooks as packs, draft changes, tests as critics, no unsupervised production. Open Writing Drafting prose with AI: audience, outline, sources, critique against the brief, human finish. No generic TED cadence. Open Case: brief to agent A worked example: take a vague request, build a pack and spec, run a short loop, and stop with evidence — without unsupervised side effects. Open Trust
3 guides One-shot gambling One clever prompt is a lottery ticket. Why first-draft fluency fools you and how to replace it with a brief, a loop, and a retry budget. Open No evaluation If you cannot fail a run, you cannot trust a run. Skipping tests turns agents into unsupervised interns. Open Secret paste Consumer chats are not your vault. What not to paste, how packs reduce leakage, and what to do if you already did it. Open Library
5 guides Glossary Bye Prompt glossary: agent, ambient, context pack, compiled spec, loop, trace, stop condition. Open FAQ Practical FAQ for Bye Prompt: is prompting dead, can I trust agents, what to automate, how this site is funded. Open Resources External starting points for agents, eval, and computer-use — plus sister field guides. We prefer primary docs over a forty-link farm. Open Methodology How Bye Prompt tests workflows, dates landscape claims, and scores tools. How to send corrections. Open Editorial Who publishes Bye Prompt, how we research agent workflows, and how to send corrections — Tabaconda LLC, Florida, USA. Open