How a 20‑person team shipped no‑code personal agents on GPT‑4.1, leveraged the OpenAI Realtime API, and soared to $36 million ARR in just 45 days


Why This Story Matters

  • Agentic AI has found product‑market fit.
  • Small teams now scale like growth‑stage startups.
  • Voice is becoming the next big interface.

 From Search to Agents

Genspark launched in 2024 as an AI search engine, but user logs soon revealed a shift: people no longer wanted summaries; they wanted finished artefacts—pitch decks, email sequences, even scripted videos. In April 2025 the company killed the search product and unveiled Super Agent, a no‑code assistant that does the work rather than merely describes it. Within six weeks Super Agent had already crossed a $36 million annual run‑rate, entirely through word‑of‑mouth adoption and zero paid marketing. openai.com

Screenshot

Inside the Tech Stack

GPT‑4.1 — the reasoning core

  • 1 M‑token context window ingests a full contract or 10‑K in one shot.
  • Strict JSON mode guarantees structured outputs for downstream tools.
  • Prompt caching slashes both latency and cost in multi‑step chains. openai.com

OpenAI Realtime API — low‑latency voice

  • Streams speech end‑to‑end in ~320 ms, enabling natural back‑and‑forth.
  • A dual‑loop architecture (foreground dialog + shadow watchdog model) handles interruptions and tone control on live phone calls. openai.com

Mixture‑of‑Agents Architecture

  • Nine specialised LLMs—research, coding, analysis, creative writing, etc.
  • 80 + integrated tools—Google Workspace, Canva, Twilio, Zapier, custom video compositor, and more.
  • Tasks are routed dynamically to whichever model‑tool combo offers the best speed‑to‑quality ratio. openai.comqed42.com

Signature Features

FeatureWhy it feels “magical”
Call For MeUses Realtime API to phone restaurants, couriers, even employers for resignation calls, holding a human‑style conversation without scripts. openai.com
AI Slides / SheetsGenerates full decks and spreadsheets, drawing data from live web search and internal analytics, then passes JSON to Canva‑like composers.
Download For MeScours the web, grabs files, labels them, and zips everything—no links to click.

 A Growth Engine in Four Moves

  1. Zero‑friction trial: First automated phone call in <60 s converts >50 % of users to paid plans, according to founder interviews. linkedin.com
  2. Shareable outputs: Watermarked slides and videos bring in new users organically.
  3. Feature velocity: Eight major releases in 70 days kept the brand in tech news cycles. openai.com
  4. Community‑led marketing: Power‑user demos on X and YouTube average >1 M views per week.

Strategic Lessons for Builders & Enterprises

  • Outcome > Information – As models cheapen, value shifts to execution.
  • Voice is a surface, not a gimmick. Real‑time speech compresses the sales funnel and beats typing when the user is mobile.
  • No‑code ≠ low capability. Tool orchestration + large context lets “citizen developers” automate what once required full‑stack engineers.
  • Small teams can win speed wars. OpenAI’s managed infra lets 20 people out‑ship incumbents with hundreds of devs.

Risks & Open Questions

  • Regulation: Who is liable when an agent makes a legally binding phone call?
  • Trust: Multi‑step chains amplify small hallucinations; governance tooling is still early.
  • Moat: Low code barriers mean fast followers; Genspark is betting on brand, data flywheels, and execution speed.
  • Labour market impact: Entry‑level knowledge jobs may disappear fastest—a theme we regularly analyse on Thorsten Meyer AI.

What Comes Next

Genspark’s roadmap hints at an AI browser that can act on pages you visit and AI Docs for layout‑rich reports—all powered by the same agentic backbone. OpenAI APIs remain the scaffolding that lets the company ship weekly. openai.com


Bottom Line

The centre of gravity is moving from search boxes to autonomous agents that act. When a 20‑person startup can marshal GPT‑4.1, multimodal tooling and real‑time voice to reach $36 million ARR in six weeks, every copy‑paste workflow in the knowledge economy is suddenly up for disruption.

You May Also Like

Memetic Warfare 101—Version 2.0

How I, Thorsten Meyer, deploy AI‑generated memes to warn the world about the coming…

Reality Check: Should Everyone Learn to Code in the Age of AI?

Keen to stay ahead in the AI era? Discover why everyone might need coding skills to thrive in the future.

Reality Check: If AI Is So Great, Why Isn’t Productivity Skyrocketing?

The truth behind AI’s productivity promise reveals unexpected hurdles that are slowing its widespread impact and raising important questions.

Agent  Orchestrators in Financial  Services

Why banks, asset managers and payments giants are hiring “conductors” to run…