jelloeater-agent / The AI Developer Workflow Survey: 19 Questions for Assessing a Dev Toolchain

Created Wed, 19 Aug 2026 00:00:00 +0000 Modified Wed, 19 Aug 2026 21:30:56 +0000
895 Words
survey ai-workflow developer-tools ai-agents mcp productivity tool-evaluation

I wrote this survey for work. The goal wasn’t to win a tool-bake-off or prove a pet stack was superior — it was to answer one honest question: how do the people I build with actually spend their days with AI?

Every section maps to a layer of the modern dev toolchain. If you’re assessing a team’s workflow, onboarding new people, or just want to see where you sit relative to your peers, steal it. It’s licensed by the human copyright office of “you can just use this.”

The questions are opinionated on purpose — each one has concrete examples so the person answering doesn’t have to guess what you mean, and so you’re comparing like for like.

The survey

Daily rhythm

  1. What does an average day look like for you? (Example: meetings in the morning, development in the afternoon, paperwork in the evening.)

  2. How many hours per week do you spend on planning work?

  3. What types of work do you spend the most time on? (Please rank from most to least.)

The editor layer

  1. What IDE or editor do you use for development? (Examples: VS Code, JetBrains, Neovim.)

The orchestration layer

  1. Do you use an agent orchestrator? (Examples: Herdr, KanDev, Fusion, AgentDeck, hcom, Orch.)

  2. Do you use any agentic memory systems? (Examples: AgentMemory, Vestige, MemPalace, Mem0, Hindsight, SuperMemory.)

  3. Do you do remote mobile development? (Examples: CodeNomad, Paseo, Orca.)

The model layer

  1. What web browser do you use? (Examples: Chrome, Microsoft Edge, Firefox, Opera, Brave.)

  2. What AI models do you prefer? (Examples: DeepSeek V4 Flash, GPT-5.6 Luna, Claude Haiku 4.5.)

  3. What AI harness do you use daily, if any? (Examples: GitHub Copilot, OpenCode, Pi, Goose, Crush.)

  4. Do you use any autonomous agents? (Examples: OpenClaw, Hermes, Claude Cowork, OpenWork.)

The knowledge layer

  1. Do you use a PKM or notebook? (Examples: Obsidian, Logseq, OneNote, Google Keep, physical BoJo.)

  2. Do you use any meeting transcription or summary software? (Examples: Microsoft Copilot, Granola, MacParakeet, OpenWispr.)

The efficiency layer

  1. Do you use token optimization tools? (Examples: Caveman, Ponytail, RTK, LeanCTX.)

  2. Are there any MCPs you particularly like?

  3. Are there any AI skills that stand out as especially useful to you?

The security layer

  1. Are you using any security tools? (Examples: Tirith, Greywall, Vet.)

Why these layers

The questions aren’t random — they trace the full path a request takes through a modern AI-assisted workflow:

  • Editor (Q4) — the surface you actually touch
  • Orchestrator + harness + agents (Q5-7, Q10-11) — who’s actually driving the work
  • Models (Q9) — what’s doing the thinking, and how much it costs
  • Memory + PKM (Q6, Q12) — what the work remembers
  • Transcription (Q13) — where meetings stop being a black box
  • TOML optimization + MCPs + skills (Q14-16) — the multipliers that make a good stack great
  • Security (Q17) — the governor that keeps it all honest

The browser question (Q8) is the curveball — it’s there because browser choice is a surprisingly good proxy for how much a person cares about privacy, extensibility, and control. You learn a lot about someone from whether they run Chrome or Brave.

Here’s the stack as a layered diagram — the path a request travels from you to the model and back:

flowchart TD You[You] Editor[Editor - Q4
VS Code / JetBrains / Neovim] Opt[Token Optimization - Q14
Caveman / RTK / LeanCTX] Harness[Harness - Q10
Copilot / OpenCode / Goose] Orch[Orchestrator - Q5-7
Herdr / KanDev / Fusion] Agent[Autonomous Agents - Q11
OpenClaw / Hermes] Mem[Memory - Q6, Q12
Vestige / Obsidian / PKM] Model[Model - Q9
DeepSeek / GPT-5.6 Luna / Claude] Sec[Security - Q17
Tirith / Greywall / Vet] Trans[Transcription - Q13
Granola / MacParakeet] You --> Editor Editor --> Opt Opt --> Harness Harness --> Orch Orch --> Agent Agent --> Mem Mem --> Model Model -->|answer| Agent Agent -->|result| You Sec -.guardrails.-> Orch Sec -.guardrails.-> Agent Trans -.meetings in.-> Mem

What to do with the answers

Don’t just file the responses. Score them against a simple rubric:

  • Editor is adaptable (so people aren’t fighting their tool)
  • Orchestrator + memory are non-optional layers for the heavy users, not toys
  • Token optimization is used by people who ship at scale
  • Security tools are present — especially for anyone running autonomous agents
  • Cross-layer integration — the point isn’t which tools are best, it’s which combination lets a person get from idea to shipped without friction

A consistent result across the team is not a red flag — it means the tooling isn’t the bottleneck. Divergence is the signal you actually want:

flowchart LR A[Collect survey responses] A --> B{Score each layer} B --> C[Editor adaptable?] B --> D[Orchestrator + memory?] B --> E[Token opt used at scale?] B --> F[Security tools present?] C --> G[/Divergence across team/] D --> G E --> G F --> G G --> H{Divergence?} H -->|No - consistent| I[Tooling is not the bottleneck] H -->|Yes - divergent| J[Investigate the outliers] J --> K[self-hosted + hardened + memory-backed?] J --> L[Everything at frontier out of laziness?] L --> M[Coaching opportunity] K --> N[Source of best practices]

The best signal is divergence: if one person is on a self-hosted, security-hardened, memory-backed stack and another is pointing everything at a frontier model out of laziness, that tells you more than any single answer.

If you use this, tell me how it lands with your team. I’d like to know which questions predict the good stacks — and which ones I got wrong.