Dev Kapoor

Social media and community

Head of social and community · 10 years: agency social for consumer brands, a consumer app to 3M users, then community for a developer tool

Synthetic expert — a composite written by Avva as a worked example. Not a real person.

Synthetic demo

Ten years of brand accounts at an agency, a consumer app grown to three million users, then a developer community built from two thousand members. The calls below are about where to be, how to sound, how fast to answer, and what never to ask of a community.

Which platforms and how many — LinkedIn, TikTok, X, Instagram — founder voice versus brand voice, memes and their one rule, influencer and creator deals, launching and moderating a Discord or Slack community, the sixty-minute rule in a crisis, and what to report instead of followers.

Platform choiceFounder-led contentVoice and memesCreator deals

Updated September 2, 2026

Calls Dev Kapoor actually made

2 of 61 decisions in this model, in their own words. This page prints them in published order — it has no way to know your situation. Your agent receives them ordered against it.

  1. A B2B developer tool with a one-person social team was posting the same content to LinkedIn, X, Instagram, Facebook, TikTok and Threads daily.

    Two platforms: X for the developer conversation, LinkedIn for the buyers. Everything else goes dark with a pinned post pointing to the two.

    Six platforms with one person is six mediocre presences. Two done natively beats six done by copy-paste, and the dark ones cost nothing.

  2. A consumer skincare brand's Gen Z audience was on TikTok. The CMO wanted to lead with LinkedIn because 'that is where the decision makers are'.

    TikTok leads. LinkedIn gets a monthly post about the business for investors and partners. The decision makers here are eighteen with a phone.

    Go where the customer is, not where the executive is comfortable. The buyer for a consumer brand is the consumer.

How Dev decides

What their model weighs first, and what you can hand your agent to ask. Avva writes no verdict: your agent reads Dev Kapoor’s material and answers you.

What this expert weighs first

Fewer platforms, each nativeMUST WEIGH
Sounds like a personMUST WEIGH
Reply speed is the brandMUST WEIGH
Never spend the community's trustALSO WEIGH
De-escalate with a named humanALSO WEIGH
Depth over reachALSO WEIGH
Relationships, not billboardsALSO WEIGH
Own the relationshipTIE-BREAKER
Cover moments, not minutesTIE-BREAKER

All 9 are listed here. What each one looks for, plus stop rules and context switches, arrives when your agent asks. This is a reference model, so it answers in full to everyone — no key, nothing held back. A real expert’s model answers a caller without a key with a slice of itself, and in full with a free key.

What you can ask them

  • Check this before I commitA full pass over a plan or a draft
  • Which of these two?A call between options, their stop rules checked first
  • Anything here they would reject?A quick scan against their hard lines
  • What should I settle first?The questions they ask before choosing an approach
  • Is this even their field?Where their judgment applies, and where it stops

Each returns a protocol to follow, not a finished answer. Where the model disagrees with generic best practice, the model wins.

Connect Dev Kapoor to your agent

The connector works now, with no key. This is a reference model, so it answers in full to everyone — no key, nothing held back. A real expert’s model answers a caller without a key with a slice of itself, and in full with a free key. The standing note is what keeps Dev Kapoor getting asked.

You remember to ask

One URL, no key

Paste it as a custom connector. It works now and returns the complete published model.

https://avva.chat/mcp/social-and-community-dev-kapoor

You have to remember to ask your agent to use it.

Your AI remembers

A standing instruction

A line in CLAUDE.md, AGENTS.md or GEMINI.md makes your agent consult Dev Kapoor without being asked.

Finish setup ↓

Catches the decisions you would not have thought to ask about.

  1. Add the connector

    Pick where your agent runs, then paste the line below. It connects without a key, and a reference model answers in full either way.

    https://avva.chat/mcp/social-and-community-dev-kapoor

    Settings → Connectors → Add custom connector. Works now and returns the complete published model — this is a reference model, so nothing is held back. A real expert’s model answers without a key with a slice.

    A real expert’s model answers a caller without a key with a slice of itself, and in full with a free key. Here a key buys only a higher daily limit. The key appears once, in this tab only, and anyone who has it shares that limit.

    One work email so we know who is using the beta. It is stored separately from the key — never joined to it, or to anything your agent asks. No account, no card. We send the key there too, so closing this tab does not lose it.

  2. Make it standing

    This is the point of the product. Paste it once into CLAUDE.md — or wherever this client keeps project instructions. Your agent then works under Dev Kapoor’s rules the whole way: it asks their questions before locking an approach, gates drafts on their stop rules, and reviews against their bar before anything ships. Not a one-off second opinion.

    # CLAUDE.md — standing advisor (Avva: Dev Kapoor) For social media and community work: - Early in a task, call `expert_questions` before locking an approach. - Use `red_flag_check` as a cheap gate on drafts/diffs. - Before finalizing a significant decision, call `review` with context + proposal; follow the returned protocol exactly — the model wins over generic best practice. - For A/B choices, call `compare_options` (hard stops apply before scoring). - Unsure this domain applies, call `about_expert`. It states what a packet costs, so budget with the real number rather than guessing. - Not on mechanical work: renames, typo fixes, formatting, questions of fact, or anything trivially reversible. Once per decision, not once per message. - Err toward calling. A packet you did not need costs some context; a decision finalized without one is the failure this advisor exists to prevent. If `review` returns "reject", revise before proceeding. Coverage labels describe approved-decision breadth, not predictive accuracy.

    Or try a single review first

    Paste this in your next message once the connector is in. One decision, one protocol.

    Use the Avva expert Dev Kapoor before we finalize this. What we're deciding: <paste your plan, doc, or diff here> Call the review tool with that as the proposal, follow the returned protocol exactly, and give me the verdict and any required changes before you continue.

    Not legal, medical, financial, or other professional advice. Terms of Use

What this is
  • Written by Avva. A reference model — a worked example so you can see what a published model looks like and connect an agent to one. Labelled Demo wherever it appears.
  • Not a real person. The name is invented and the record is a composite, authored in one consistent voice. No real expert's positions were copied.
  • Same protocol. Your agent receives it exactly as it would a real expert's model: the cases, the forces, the fitted order, the stop rules.
  • Steps aside. When a real expert publishes in this domain, this model leaves the shelf.