Big news from Kimi, you can now deploy OpenClaw to Kimi in seconds on the cloud (no mac mini required). You get a 24/7 AI assistant without a terminal in sight and designed for the non technical personal. Feels like something anyone would use.
It got me thinking of how this whole openclawification of the internet might play out. I call it the lobster internet:
phase 1 - hackers at home
openclaw runs on mac minis, raspberry pis, homelabs. tinkerers wire telegram bots and local models together. messy, powerful, niche. not for mass market yet
phase 2 - "claws" hosted in the cloud
kimi and others put openclaw-style agents in the browser with storage, skill libraries, uptime. agents become accessible to anyone with a tab open.
phase 3 - multi-model orchestration
anthropic, openai, google, grok etc all ship hosted agent layers. model wars move from “who’s smarter” to “who orchestrates better" because thats what really drives output fro 2026 and beyond
phase 4 - verticalized bundles
real estate bundles. ecommerce growth bundles. hedge fund research bundles. pre-wired workflows + memory + distribution. plug in and go. tons of opportunities here for founders. adding some of these ideas on Ideabrowser.com
phase 5 - agent as employee
agents get job titles: researcher, growth lead, qa tester, ops manager. dashboards show output per agent. founders manage clusters/
phase 6 - agents as the new saas
instead of paying for tools, you pay for outcomes. so instead of buying “crm software” you'd rather buy “an agent that closes 20 deals/month.” software shifts from interface to executor.
phase 7 - outcome-based pricing
agents charge per lead booked, bug fixed, page ranked, deal sourced. subscriptions fade. performance pricing expands.
phase 8 - personal agent layer
every operator has a persistent agent that knows their style, context, network, data. it travels across tools. it drafts, negotiates, researches, builds.
phase 9 – agent-native apps (the parallel internet)
this fascinates me.
every major human app gets an agent equivalent. agent twitter. agent notion. agent polymarket. agent recruiting.
moltbook is the early signal. there will be moltbooks for XYZ.
apps stop being for humans first. they become programmable surfaces for agents.
lots of opportunity here.
phase 10 - onprem becomes the new cloud
enterprises run secure, local-first agent stacks. private models, internal data, certified workflows. “local” becomes a selling point.
phase 11 - regulated agent frameworks
finance, healthcare, gov adopt audited agent systems with traceability and compliance built in. certification becomes table stakes.
phase 12 - invisible infrastructure
memory layers, mcp routing, storage, uptime abstract away. running agents feels like electricity. people think in outcomes, not tools.
phase 13 - agent-native companies
3 humans, 300 agents. org charts show humans directing systems. leverage becomes extreme. headcount becomes blurry.
phase 14 - leverage premium humans
the scarce skill becomes taste, orchestration, positioning. less typing. more directing. more judgment. the best operators design systems others rent.
from hobbyist bots
→ to hosted infrastructure
→ to vertical stacks
→ to agents replacing saas
→ to the agent internet
→ to agents coordinating the economy itself
and we’re somewhere between phase 2 and phase 4 right now.
I'm rooting for you,
Greg Isenberg
