AI-native assessment — Prasad Gupte

KnowledgeContext

Where does the knowledge your AI works from live?

0 Personal chats Looks like: the brief pasted into ChatGPT, Gemini or Copilot again — and gone by next week. 1 Shared chats & artefacts Looks like: shared ChatGPT or Claude chat links, a Slack or Teams thread of the prompts that worked. 2 Projects with attached context Looks like: ChatGPT or Claude Projects with the brief, brand guide and Notion or Confluence pages attached. 3 One context shared across projects & teams Looks like: one knowledge base — Notion, Confluence or SharePoint — every project connects to, with an owner.

ToolsHarness, skills

What can your AI use beyond the chat window?

0 Chat only Looks like: ChatGPT or Copilot in a browser tab — paste in, copy out, nothing connected. 1 Saved prompts & personal assistants Looks like: a saved prompt, a Custom GPT or a Gemini Gem that one person built and owns. 2 Shared skills anyone runs by name Looks like: a shared Claude skill or team GPT anyone runs by name — same output, whoever asks. 3 Skills wired to your systems Looks like: skills connected to Jira, HubSpot, Shopify or the repo through connectors — no copy-paste.

AutomationAgents & workflows

What runs without someone starting it?

0 Nothing Looks like: every prompt typed by hand — nothing runs on a schedule, nothing runs overnight. 1 Rule-based workflows Looks like: Zapier, Make or n8n flows, scheduled Gemini or Cowork jobs, Power Automate drafting the reply. 2 Agents on demand, schedule or events Looks like: an agent that fires on a new support ticket, a calendar slot or a new row in the sheet. 3 Agent pipelines that hand off Looks like: a pipeline that hands the PRD to the design agent, then the design to the repo.

CollaborationShareability

How far does AI-made work travel?

0 Pasted into a doc or slide Looks like: AI output pasted into Google Docs, a PowerPoint or a Slack screenshot — and it stops there. 1 Shared artefacts & touchable prototypes Looks like: a Claude artifact or ChatGPT canvas link, a Lovable or Figma Make prototype others can click. 2 Shared projects others build on Looks like: one shared Lovable, Replit or Figma project the team keeps extending — not one author's copy. 3 Merged into production Looks like: Cursor, Copilot or Claude Code changes merged through a pull request and live for customers.

CostToken spend

How is AI spend governed?

0 Unmetered usage Looks like: ChatGPT Plus on expenses, API keys on someone's card — nobody adds up the total. 1 Overall budget Looks like: a Cursor or Copilot seat count and one AI line in the monthly budget review. 2 Individualised budgets Looks like: a token budget per team in the OpenAI or Anthropic console — each team picks its models. 3 Cost per outcome Looks like: cost per resolved ticket, per product page or per merged PR — and contracts priced on it.

QualityEvals

What stands between the model and what ships?

0 Depends on the prompt Looks like: the draft goes straight into the newsletter, the product page or the customer reply. 1 Human in the loop Looks like: an editor signs off the copy, a team lead approves the reply, a reviewer reads every PR. 2 Critique / guardrail skills Looks like: a brand-voice checker, a claims and GDPR check, an AI code review — before a person reads. 3 Eval suites Looks like: 50 real tickets or briefs with known-good answers, re-scored on every prompt or model change.

PeopleTeam roles

How has the team shape changed?

0 Same org chart Looks like: the same PM, design and engineering hand-offs as before — AI stays a personal tool. 1 T-shaped individuals Looks like: a PM who prototypes in Lovable, a marketer who builds her own n8n flow, a designer who ships code. 2 Smaller cross-functional pods Looks like: a pod of three owning checkout end to end — work that used to take eight people. 3 Roles rewritten (JD blocks) Looks like: job descriptions and career ladders with AI-native blocks, and hiring loops that test them.

ValueOutcomes

What do you measure?

0 Anecdotes & satisfaction Looks like: “it saves me hours” at the all-hands, a pulse-survey score, a few champions' stories. 1 Adoption & usage Looks like: weekly active users in the ChatGPT Enterprise or Copilot dashboard, seats used vs. bought. 2 Productivity & impact Looks like: cycle time in Jira, tickets resolved per agent, campaign assets shipped per week. 3 Outcomes tied to P&L Looks like: support cost per order, conversion on AI-written product pages — the P&L line that moved.

How you run

How people get enabled, and what's already automated in production.

09 How do people get enabled?

1 Self-taught Whoever is curious figures it out. 2 Off-the-shelf training Generic courses, licences, lunch-and-learns. 3 Outcome-based training on our own backlog People learn by shipping items from the real backlog.

10 What use cases have already been automated?

Running in production, not piloted.

About your organisation

Role, industry and size are required — they're how the room gets compared. Company is optional; we e-mail you the result.

14 Company (optional) optional

15 Where should we send your Octagon?

We e-mail you this result once, with a link to revisit it. Stored with your answers. No newsletter.

Where are you on Builder Impact?