The Plugin Gold Rush: ChatGPT Just Opened the Biggest Free Distribution Channel of 2026

The Plugin Gold Rush: ChatGPT Just Opened the Biggest Free Distribution Channel of 2026

Greg Isenberg says he can’t think of a better risk/reward bet right now than building a ChatGPT plugin. Not a startup. Not a SaaS. A plugin. Because as of yesterday, ChatGPT started recommending plugins in the middle of conversations — and millions of questions get asked every day with no plugin to recommend. Every one of those questions is an empty slot, and the slots are free. For now.

The mechanism is what matters. This is not the app store with a search ranking. ChatGPT sits inside the user’s intent at the exact moment of asking — the plugin is recommended by the conversation itself. That collapses the entire acquisition funnel: no SEO, no ads, no content marketing. The user asks, the assistant suggests your tool, your tool does the thing. @skirano, quoted by Isenberg, claims his plugin extension is up over 2,000% since the announcement.

Isenberg lived through one of these windows himself, building venture-backed companies in San Francisco, and he calls them “sacred times”. His playbook is deliberately unglamorous: find a tool or workflow people already Google for or do by hand every week — PDF to Excel, invoice generators, weekly client reports — and build it as a plugin. Then write the description using the exact phrases people type into ChatGPT. That detail is the whole game: it’s SEO for conversations. The plugin description is the new meta title. Ship five of them, watch which ones ChatGPT picks up, and double down on the winner.

The pick-and-shovel signal

The most telling reply came from @ideabrowser: “Literally giddy about this App Store moment. Adding a ‘plugins ideas’ section to ideabrowser.” When the idea marketplaces start building for the wave, the wave is real. Ideas are the inventory of a distribution land-grab — curation becomes a business.

• Conversational recommendation is a fundamentally new distribution primitive: the assistant sells the tool at the moment of intent.

• Plugin descriptions are the new SEO — written in the user’s own query language, not the marketer’s.

• Free-distribution windows are short (months, not years) and look obvious only in hindsight; Isenberg puts the close at ~12 months.

• The highest-leverage move is boring: digitise a weekly chore people already do by hand.

Read the original post → · ideabrowser’s reply →

The Vertical Agent Manifesto: Generic Assistants Lose, Deep Products Win

Luke Ramsden’s thesis is crisp enough to fit on an index card: consumer AI will not be won by iMessage-style “personal assistants” or generic horizontal agents. It will be won by high-polish products that focus on a specific area of a person’s life, build proprietary context on the person, and use that context to create a better marketplace or network than existing pre-AI platforms can.

Notice the three-part machine inside the sentence. Focus (one life area) produces proprietary context (data no horizontal assistant has, because it never goes deep enough), and that context is then spent on a better marketplace or network — the moat is not the model, it’s the matching. A generic assistant can answer questions about your fitness; a vertical product knows your injuries, your schedule, your gym, and can match you with the right trainer better than any directory. That matching layer is where the defensibility compounds.

This sits in productive tension with the lead story. ChatGPT plugins push toward one horizontal everything-app with 1.2 billion users; Ramsden argues the durable consumer winners are vertical. Both can be true: plugins are the distribution layer, vertical products are the destination layer. The plugin gets you in front of the user; the proprietary context keeps them. It also rhymes with Edition 3’s Harsha Kumar essay on the AI assistant wars — the “family butler” endgame is vertical by definition.

• Depth beats breadth in consumer AI: proprietary context on one life area is the moat, not model quality.

• The moat expresses itself as a marketplace or network — better matching than pre-AI platforms, powered by context.

• Horizontal assistants risk becoming a commodity distribution layer that vertical products ride on.

Read the original post →

Trust Is the Feature: Fo’s Humans-in-the-Loop Bet on Agents

Vinod Khosla’s verdict on the agent race: the market is far from decided. His evidence is Fo, a new personal AI introduced by Shivani Poddar — who led engineering at Google DeepMind — that refuses the industry’s favourite pretence. “Other personal AIs pretend AI can do everything,” she says. “Fo employs humans to do tasks that AI cannot.”

The mechanism is a trust unlock. Agents stall at the exact point where the stakes get real — credit cards, bookings, money movement — because a 95%-accurate agent is a 5%-disaster agent. Khosla’s claim: Fo is “more trustable, safer with credit cards, much higher completion rates,” and humans in the loop are what solve the harder tasks. The numbers he cites: two times better at real-world task completion than the alternatives.

This is the pragmatic counterpoint to the “agents eat everything” maximalism that has run through recent editions. A human fallback is usually framed as a crutch — a confession that the model isn’t good enough. Fo reframes it as architecture: the human is the trust layer that lets the agent take on consequential work at all. More trust means more permissions, more permissions mean more value, and the human-handled edge cases are the training data that eventually automate them away. The loop is the product.

• The agent bottleneck isn’t capability, it’s permission: users won’t hand over credit cards to a 95% system.

• Humans-in-the-loop is a trust architecture, not a stopgap — it converts unusable autonomy into usable completion rates.

• Every escalated task is free training data; the human layer is a flywheel, not a cost centre.

Read the original post →

The Palace of Gold, Now in ATM Form

Kalyan Jewellers opened its first ATM store in Chennai last month, and yes — your first reaction is the correct one. ATM here stands for Akshaya Thanga Maligai, “The Palace of Gold” in Tamil. Kalyan has committed ₹300 crore to build five such stores in Tamil Nadu, aimed squarely at the ₹16,000-crore regional giant Thangamayil.

The strategy underneath the gimmick is format innovation against an entrenched incumbent. Thangamayil’s moat in Tamil Nadu is trust built over decades — in gold retail, the brand is the assay certificate. You don’t out-trust a ₹16,000-crore local giant head-on. So Kalyan is changing the shape of the store instead: the ATM format means smaller footprints, lower capex per location, faster rollout, and the novelty of buying gold the way you withdraw cash. Five stores for ₹300 crore is ₹60 crore a store — a fraction of a flagship — which buys speed and presence in a market where being nearby is half the sale.

• In trust-moat categories like gold, frontal assault fails — format innovation is the flanking move.

• Smaller footprints buy speed: five quick openings can matter more than one flagship.

• The name does real work: “Palace of Gold” in Tamil signals this is built for the local buyer, not imported.

Read the original post →