Soloop.io - AI team - 2026 - EN

Материал из Wiki - Iphoster - the best ever hosting and support. 2005 - 2026
Перейти к:навигация, поиск

Published: 2026-09-04


What Soloop Is and Why It Exists

Soloop (soloop.io) is an Agent OS — an operating system built around AI agents — designed for one specific audience: solo founders. Not teams, not enterprises, not agencies. One person with a product and no team. The tagline on the homepage does the heavy lifting: "Idea In. Revenue Out" and "Make your product make money."

The core premise is that a solo founder should not have to hire a company to run a company. Instead, Soloop supplies a virtual executive team: an AI CEO that plans, an AI CTO that ships code, an AI CMO that finds users, an AI Analyst that reads the market, and an AI CFO that answers finance questions. Each agent has a durable role — not a one-off prompt, not a chatbot persona that resets with every message.

What separates Soloop from most autonomous agent products is the approval-first architecture. Agents do not publish, send, code, or spend anything without explicit sign-off from the founder. Every meaningful action is packaged into a single card: objective, exact action, target, credit cost, and stop condition. The founder reads the card and chooses one of three paths: approve, request changes, or cancel.

The product launched in early August 2026, took second place in the Product of the Day ranking on Product Hunt, and quickly became a reference point in the "one person plus AI team" niche. It is young, small, and honest about being both.

The Core Problem: Demos Got Cheap, Demand Stayed Expensive

Soloop's philosophy rests on one sentence from its own documentation: "Demos got cheap. Demand stayed expensive."

In 2026, AI coding tools made the cost of producing a working product collapse. One person with Claude Code, Cursor, or a similar tool can build an MVP in an evening. A landing page is an afternoon. A demo is a weekend. The bottleneck moved — it always does — and it landed exactly where the tools cannot help: choosing an audience, finding real users, reading their feedback, deciding what to build next, and ultimately converting attention into revenue.

The founder, Wenhao Yu, who writes publicly as Arthur Yu, describes the pain from direct experience at Baidu:

  • "At Baidu, I saw tons of demos get stuck because the teams didn't know how to iterate next."
  • "In the end, they didn't make a single penny."
  • "I hope Soloop ends it."

This is why Soloop deliberately starts at step two, not step one. The platform assumes the founder already has something on the table: a GitHub repo, a landing page, a deck, a demo, a ZIP archive, a half-built product, or a stubborn insight. From there, Soloop helps choose the next move, asks for approval, and sends the right agent to do the work.

It also explains the most distinctive number in the product: ten returning users, then the first dollar. Until that milestone is reached, no new feature deserves development. A user who comes back is worth more than a waitlist signup. A product is not real until money moves.

The Team Behind Soloop

Soloop was built by a small founding team with Chinese and international background, several members of which literally crossed the table from investor to founder.

Founder and CEO: Wenhao Yu (Arthur Yu)

Wenhao Yu is the founder and CEO. On the homepage he is credited as Wenhao Yu; in the About section he writes as Arthur Yu. It is the same person. He is based in Singapore, and his X profile (@ArthurYu_Soloop) reads: "Building soloop. Ex-Baidu MeDo, YC China (F24). The world has enough demos."

His background includes Baidu (the MeDo division) and YC China (batch F24). At Baidu he watched teams solve hard product problems while drowning in the decision pipeline: meetings, quarterly reviews, reports, handoffs. When AI coding tools later let one person ship a working product alone, the remaining gap was the work code cannot cover — audience selection, user acquisition, feedback reading, and the question of what to build next. Soloop is his answer.

CFO: Qingfeng Meng

Qingfeng Meng is a co-founder and CFO. He first encountered the company as an investor. At Shuimu Tsinghua Alumni Seed Fund he worked on investments in more than thirty Tsinghua-founded startups; later he ran global go-to-market at Zhipu AI. After the fund invested, he crossed the table and joined the founding team.

His frame is venture thinking: every important decision is an allocation of capital, where capital includes cash, the founder's time, and the paths foregone by choosing one direction over another. "Building, selling, and waiting all have a cost." He measures an AI team by how much uncertainty it removes each week. His sharpest formulation: most AI tools make a single task cheaper; Soloop aims to reduce the cost of deciding what the company should do next.

CTO: Xinyu Shi

Xinyu Shi is co-founder and CTO. His core lesson from building Soloop: a useful AI team is very different from putting several agents into the same product. The hard part is making every agent understand the same company goal, work from the same context, and hand its output to the next agent in a usable form. Soloop's technical pillars are shared memory, clear ownership, reliable handoffs, execution environments, approval checkpoints, and feedback loops.

Growth and Brand: LangDavidDai

LangDavidDai is co-founder responsible for market, brand, and commercialization. Before Soloop he worked in media and platform marketing plus creator operations, and co-founded a Gen Z startup at MiraclePlus (batch 2020S). His positioning line: "Vibe coding gave one person a software team. Soloop is building the rest of the company. You keep the vision and final say."

Legal and Operating Details

The operator is Soloop, Inc., registered in the United States (Newark, California) with Delaware governing law. Contact email is admin@soloop.io. Product Hunt also lists Zac Zuo as the launch hunter.

The team openly says the product is early. They respond to criticism publicly, take community suggestions into the official design focus, and dogfood their own product — Arthur Yu admitted in comments that the team uses its own agents to draft posts and approves everything themselves.

Product Philosophy and the Ten-User Milestone

Soloop is rare in its niche because it publishes a point of view. The principles that guide product decisions are printed on the site:

  • "A product is not real until money moves."
  • "Demos are cheap now. Demand is still hard."
  • "One clear next move beats ten vague options."
  • "User proof beats founder conviction."
  • "A solo founder can build faster than they can learn."

The operational target is blunt: ten users who come back, then the first dollar. The loop that matters is: find real people, hear what blocks them, change the product, decide what deserves another week. Features earn their place only if they move that loop. It is a discipline that most startups never adopt — they build features endlessly and collect no evidence of demand.

On the future, Soloop quotes Sam Altman (via Alexis Ohanian's post): "There is a one-person billion-dollar company — which would have been unimaginable without AI and now will happen." Soloop's banner is "1 person, $1 billion." But the team is explicit about the boundary: they believe in one-person companies, not zero-person companies. Judgment stays human. The founder remains the last decision-maker.

How Soloop Works: The Five-Step Loop

The product manual describes a five-step loop, and the design discipline is visible at every step.

Step 1: Context

The founder brings what already exists: an idea, a URL, a GitHub repo, a ZIP (up to 50 MB), PDF, DOCX, an image, a pitch deck, a screenshot, or a half-built product. Up to five files per message.

Step 2: One Call

The AI CEO reads the project, checks active work, and picks exactly one useful next move. Soloop never offers ten options. It offers one card, one reason, and one way to judge whether the work was worth doing.

Step 3: Review

The founder sees the exact card: objective, exact action, target, credit cost, and stop condition. Coding cards add acceptance criteria — a definition of done visible before work starts. The three choices are Approve, Request changes, or Cancel. Cancel closes the proposal; nothing runs.

Step 4: Work

The approved work is executed by the matching specialist: market research, a post on X, a cold email, a code change in the sandbox, or a deploy preparation.

Step 5: Signal

The result changes the next call. Replies, users, code changes, task notes, and failed attempts stay in the workspace and shape the CEO's next decision.

One important distinction from the documentation: there are two kinds of cards. An action card starts concrete work. A handoff card opens a specialist workspace (CMO, CTO, Analyst, CFO) — approving it does NOT run external work; the specialist must first prepare a separate action card. This prevents accidental execution during context transfer.

The Agents in Detail

AI CEO: The Planner and Coordinator

The AI CEO is the center of the system: main conversation, project view, next-move selection, and routing. It explains decisions and does not execute external actions itself — an architectural choice, not a limitation. The CEO turns conversation into tasks for the Analyst (market and users), the CTO (build), and the CMO (users and feedback).

Competing priorities between agents are handled without a meta-agent. CFO Meng's answer: competing priorities are a visibility problem, not an AI problem. Context handoffs between CEO, CTO, and CMO are visible to the founder by default — who is working on what, what was handed off, what is waiting. "You don't pick an agent because of their title. You pick the agent whose thread you're currently following." His internal test: how much uncertainty does the founder have about what their own company is doing right now? The less, the better.

AI CTO: The Builder in a Sandbox

The AI CTO is the agent with teeth — the only one that actually changes code. Approved jobs run in an isolated sandbox and can return a checked change or a Vercel deployment. Input: a new project, a GitHub repo, or a ZIP up to 50 MB.

Coding cards carry acceptance criteria so "done" is defined before the agent starts. For non-technical founders, approval is not a code review. The team explains that approval checkpoints are translated into plain-English business trade-offs: "This architecture gets us to launch two weeks faster but limits custom features in month three. Proceed?" As CFO Meng puts it: "You don't need to understand how it's built. You need to understand whether to keep going."

AI CMO: The Demand Generator

The AI CMO turns attention into demand. As of September 2026, channel statuses are:

  • X / Twitter — live with OAuth: drafts posts and replies, publishes approved work, collects comment signals (three-day comment watch).
  • Cold email — beta, Gmail required: prepares an operating plan and single-recipient send or reply cards; sending follows approval mode.
  • Reddit — manual publish: prepares a reviewable post or reply and opens a prefilled Reddit flow; publishing stays in the founder's hands.
  • Ads — plan only: the CMO can discuss the channel and put it in a plan; it does not buy or manage ads.

The team admits the CMO feedback loop is the longest and noisiest. Commenter Michael Vavilov put it precisely: product iteration (CTO layer) is easiest to trust because code has fast measurable feedback loops; CMO decisions are hardest to trust because audience choice, messaging, and distribution produce slow, noisy signals that are hard for an agent to learn from.

The counter-argument from LangDavidDai is the strongest part of the exchange: distribution noise is exactly why solo founders need AI help there more, not less. A one-person team cannot run ten messaging experiments, analyze five channels, and synthesize what works — you cannot even hear the signal because you are drowning in execution. The AI CMO runs the experiments, surfaces the patterns, and brings the signal so the founder can make the call in minutes instead of weeks. Human judgment on positioning and audience remains the founder's.

AI Analyst: The Researcher

The AI Analyst runs approved competitor, market, and target-user research. Reports stay on the task and include source links when collected. Documentation suggests useful first requests: compare five competitors and show evidence before naming the gap; find ten recent X posts where solo founders complain about getting first users; inspect a repo and prepare a task to fix signup drop-off with acceptance checks.

AI CFO: The Advisor That Does Not Move Money

The AI CFO is conversation-only. It works through pricing, cost, and finance questions, but does not move money and has no separate finance worker. It helps the founder think about unit economics.

This matters for an honest evaluation: of the five roles, only three actually execute — CTO, CMO (partially), and Analyst. The CEO only plans and routes; the CFO only talks. Independent reviewers (MakerStack) call this out bluntly: "two of five agents cannot execute anything."

Channels: Live, Beta, Manual, Plan-Only

The manual publishes channel status honestly, with the rule: "Social means Twitter/X today. Reddit, email, ads, and other channels stay off this page until there is runtime behind them."

Channel Status What it does
Analyst Live Approved competitor, market, and prospect research; reports with source links
X / Twitter Live with OAuth Drafts posts and replies, publishes approved work, collects comment signals
Cold email Beta, Gmail required Operating plan and single-recipient send or reply cards after approval
Coding Live Work from a new project, GitHub repo, or ZIP; sandbox, Vercel deploy
Reddit Manual publish Prepares a reviewable post and opens a prefilled Reddit flow
Ads Plan only Discusses the channel and includes it in a plan; never buys or manages ads

The honest summary: one fully automated public channel (X), one beta channel (email), one assisted channel (Reddit), and one conversation (ads). For a product pitched as "delegate acquisition," that gap matters. What deserves credit is the honesty: the limits are printed in the documentation, not hidden behind marketing.

Approval-First: The Card as the Unit of Work

What a Card Contains

The approval card is the whole product and, per many reviewers, the best thing in it. Any action that posts publicly, spends credits, runs research, changes code, or deploys first generates a card containing:

  • Objective — why this is being done.
  • Exact action — what will happen.
  • Target — how to tell if the work was worth doing.
  • Credit cost — what it costs.
  • Stop condition — when the agent must stop.
  • Acceptance criteria — for coding tasks.

Most agent products bury this behind a confirmation dialog. Soloop makes the card the unit of work. That is a real product difference, not a slogan.

The Trust Ladder

The question every approval-first product faces: how do you stop approval from becoming a habit, and the founder from becoming a rubber stamp? Peter Digitalis put it sharply on Product Hunt: "A founder who approves everything is reading nothing." And: "If declining has no visible consequence, the yes was never a decision."

Soloop's answer is the trust ladder, articulated by CFO Meng:

  • Reversible, low-stakes actions (drafts, routine reporting, post preparation) shift toward "execute and log" without separate approval.
  • Irreversible, high-stakes actions (capital expenditure, publishing production code, launching ad spend, changing pricing) require hard sign-off.
  • The threshold moves as confidence metrics grow: more clean runs, lower threshold.

LangDavidDai described the structural measures against "approval theater":

  • Risk-tiering — not all approvals are equal: a landing page copy tweak is low-stakes, batched daily and shown as a diff; a pricing change or new feature direction is high-stakes, surfaced immediately with full context.
  • Diff-only approvals — show what moved and why, not the whole document.
  • Batching by default — the default is not "ping me for everything" but "here is everything that happened since you last checked, organized by what needs your call." The founder sets the rhythm, not the AI.

The team is candid: "This is still the hardest design problem we're working on, and we haven't cracked it perfectly."

Novelty, Time, and the Cost of Saying No

The Product Hunt community contributed two refinements that the team officially took into design focus:

  • Novelty-tiering (Tiffany Trboyevich): any new task type gets full review until it produces two consecutive clean rounds, regardless of stakes. Stakes tell you how bad a miss would be; novelty tells you how likely one is. Without both axes, a familiar-looking failure walks right through the low-stakes door.
  • Time-tiering (Hugo Rodrigo Baigorria): the threshold should move with elapsed use, not only with confidence metrics. And crucially — decouple visibility from the gate. Visibility should stay high forever; the founder should always be able to read what agents did. The gate is what should fall away, and fast. Tying them together means loosening approvals also removes the narration, and things start to feel like they are happening behind your back — which costs more trust than the approvals ever bought.

And the "what happens if I say no" question: agents should state the consequence of declining. "Declining means we pause this acquisition channel and save $500 but lose X leads" turns approval from a passive green button into an active trade-off. Soloop announced that a stakes-dynamic UI is officially in the design focus.

Memory and Documents: Product Doc, Tasks, Inbox

Soloop does not treat chat as the source of truth. Chat is one view. The durable record lives in four places:

  • Product Doc — the editable brief for the product, target user, problem, advantage, and current focus.
  • Uploaded documents — files stay with the project; supported text documents are indexed, with visible indexing status.
  • Tasks — the source of truth for pending approvals, running work, failures, results, and execution notes, including credit use.
  • Inbox — external replies and monitored signals that may deserve a response or a change in direction.

When asking about prior work, the docs recommend naming the file, task ID, reply, or Doc field so the CEO loads the right record. It is a deliberate design against the "scroll up and hope" pattern.

File Limits

Type Formats Limit
Documents TXT, Markdown, PDF, DOCX up to 10 MB per file
Images PNG, JPEG, WebP, GIF up to 3 MB per image
Attachments up to 5 files per message or landing-page start
Existing code GitHub repo or ZIP up to 50 MB
Long text text attached to a chat message up to 50,000 extracted characters per file

Troubleshooting per the manual: if a card is approved but the task did not start, check the task state — common holds are a missing account connection, exhausted credits, provider quota, or another coding job using the project sandbox. Background results do not always return to the open chat; look in Tasks, Documents, and Inbox.

Pricing and Credits

Soloop uses a monthly credit balance that funds all work: planning, research, coding, and growth. Credits refresh monthly and do not roll over. Pricing as of September 2026:

Plan Price Credits Includes
Free $0/mo 15 credits Core planning and chat, review every output before use, help center, no credit card
Builder (most popular) $39/mo 200 credits Research, coding, and growth workflows; approval-first external actions; standard support
Operator $99/mo 600 credits Everything in Builder, recurring operating runs, monitoring and approval queue
Team from $199/mo 1,500–7,500 credits (1500/3000/4500/6000/7500) Everything in Operator, early access, priority support

Annual billing takes 15% off. The trial is seven days of Builder with 100 trial credits (card required); without cancellation it converts to Builder at $39/mo. Payments run through Stripe, including hosted checkout and the billing portal.

The Pricing Criticism: A Credit Is Never Defined

The sharpest criticism from independent reviewers (MakerStack, 6.5/10) is that a credit is never defined. Nowhere on the pricing page, the docs, or the about page does Soloop publish what a credit buys. There is no rate card, no "a research run costs roughly X credits," no worked example. A founder choosing between $39 and $99 cannot tell whether 200 credits is forty research runs or four. The card shows the cost — after you have already paid for the month. The free tier's 15 credits of an undefined unit answers nothing.

This matters precisely for Soloop's target buyer. Reviewers suggest treating Builder as a two-week plan rather than a monthly one, and the free tier as a metering experiment rather than a trial. Until a credit rate is published, the economics are unpriceable.

Competition: The One-Person Company Wars

The niche is crowded. Soloop is neither the first nor the only player.

Paperclip: The Open-Source Orchestrator

Paperclip (paperclip.ing, GitHub repo paperclipai/paperclip, MIT, roughly 80,000 stars) is the open-source control plane for teams of agents. The formula: "If OpenClaw is an employee, Paperclip is the company." A Node.js server plus React UI that organizes agents into an org chart — roles, reporting lines, budgets, governance, ticket system, heartbeats, cost control with hard stops.

Key differences from Soloop:

  • Paperclip is a kit of primitives, not a product: configuring org structures, prompts, and integrations takes hours to days.
  • Bring-your-own-agent: hire Claude Code, Codex, Cursor, OpenClaw, Bash, or HTTP bots. Soloop supplies ready specialists.
  • Paperclip is free and self-hosted, but you pay for models and hosting. Soloop is paid SaaS with a ready loop.
  • Security: in August 2026, Oasis Security disclosed critical flaws, including CVE-2026-41679 (CVSS 10.0) — unauthenticated RCE via self-registration plus importing a .paperclip.yaml with a process adapter. The conclusion: "an agent config becomes executable behavior." The patch shipped in v2026.416.0.

Community sentiment on Paperclip is mixed: praise for the framing ("CEO hires Coder, you approve"), complaints (GitHub issue #3308) about weak agent coordination, unclear roles, and stopping at planning without implementation — exactly the problems Soloop targets with one card instead of background soup.

Tycoon AI: The Closest Direct Rival

Tycoon AI (tycoon.us) launched May 21, 2026 and took #1 Product of the Day (474 upvotes, 111 comments). Founder Xiaoyin Qu. Key elements: an AI CEO named Astra, ten-plus ready agents (CMO, CTO, CFO, COO, Head of Content, Research, SEO, support), a "Company Brain" memory, Decision Cards with a "Needs You" queue, and routine automation (claimed: up to 65% of tasks automatically).

Differences from Soloop:

  • Tycoon is a "company from the box" in thirty seconds. Soloop says "bring what you already have," with focus on finding users and the first dollar.
  • Tycoon has more roles and multi-channel breadth; its memory is a system of documents, rules, and lessons.
  • Pricing is a usage wallet from $50/mo: tokens at OpenRouter prices with no markup, machine-hours (1 GB $0.032/hr, 4 GB $0.106/hr), and infrastructure (domains, email, $3/GB storage).
  • Criticism (MakerStack, 6.3/10): the pitch that "AI runs the whole company" is overstated — agents are uneven, strategy still needs you; in 2026 this is a strong assistant, not an autonomous CEO.

Polsia: Full Autopilot (The Antithesis)

Polsia (polsia.com) is the philosophical opposite. Founder Ben Broca builds fully autonomous companies: the CEO instance wakes at night, assesses state, decides, executes, and sends a morning email. It provisions everything itself — email, Render, Neon, Stripe, GitHub repos. "Provision everything, ask nothing."

Claimed numbers are striking: over 1,000 companies, $1M ARR in about thirty days (per Latent Space), a $30M round at a $250M valuation, roughly $10M ARR by August 2026, zero employees. Price: $49–50/mo plus 20% of revenue (incentive alignment).

The difference from Soloop: total autonomy versus approval-first. Soloop and Tycoon both position themselves against Polsia — control and transparency beat autopilot. Criticism of Polsia: all figures are self-reported and unverifiable; questions of accountability (who answers for bad ads?), service dependency, and inference cost; Broca himself admits a retention challenge.

Other Players in the Niche

  • Lindy (lindy.ai) — no-code AI teammate in Slack and email: agent steps and triggers, 1,000+ integrations; $30/$100/$200 per user. Closest commercial analog on price, but no C-level roles — task automation, not a company.
  • Bond (bondapp.io) — "AI Chief of Staff": reads Gmail, Calendar, Slack, Notion, Linear, ranks tasks, drafts; executes nothing autonomously. #1 on Product Hunt (665 upvotes).
  • DayZero — blueprint engine turning ideas into strategic plans; separately, dayzero.ai sells "AI Employees" for B2B. Not a competitor on the idea-to-revenue cycle.
  • Taskade — all-in-one workspace with multi-agents; general productivity tool.
  • Devin / OpenHands / Amp / Claude Code — autonomous coding agents (one role: engineer). An alternative path, not direct competition.
  • Naïve — infrastructure for the "body of a company": incorporation, email, memory, compute via one API; $28.5M Series A, 30k developer customers. Complementary, not a competitor on UX — but it eats the bottom layer of the niche.

Comparison Table

Product Positioning Roles/agents Price Key difference from Soloop
Soloop Approval-first Agent OS, "idea to revenue" CEO, CTO, CMO, Analyst, CFO Free / $39 / $99 / from $199 One approval card as the unit of work
Paperclip Open-source control plane for agent teams Any (BYO): CEO, engineers, OpenClaw, etc. Free (MIT) plus model costs A kit of primitives requiring setup; Soloop is a ready loop
Tycoon AI "Company from the box" Astra (CEO) plus 10+ agents Wallet from $50/mo More roles and automation, but no approval cards
Polsia Full autopilot CEO plus Eng, Marketing, Ads, Support $49/mo plus 20% of revenue Autonomy versus approval-first
Lindy AI teammate in team workflows Workflow agents $30–200/mo No C-level roles and no single company goal
Bond AI Chief of Staff One agent Early access Prioritization and drafts only, no execution

Vibe Coding vs. an AI Team: Why Many Choose Claude Code

Understanding Soloop's position requires understanding why most solo founders chose a different path. The original Telegram post that kicked off this research (channel "В эпоху AI," t.me/create_products/754) is telling: the author built an AI company on Paperclip, discovered it was expensive and not so simple, and returned to vibe coding with Claude Code. The arguments are systemic:

  • Cheaper and simpler: the standard 2026 stack — Cursor $20, Claude Code $20, Replit $25 — comes to about $65/mo, versus $39–199/mo for agent platforms and $300–500/mo for "full agent stacks."
  • Direct control: the founder stays the CTO, sees diffs, and avoids a coordination layer of "tickets from background soup."
  • Trust in your own eyes: 63% of vibe coding users are not developers; Karpathy's framing ("I state intent — the LLM writes"); proof points like Pieter Levels ($3M+, zero employees) and Marc Lou (ten-plus micro-SaaS).
  • Cutting the worst class risks: no approval fatigue, no opaque credits or wallets, no vendor lock-in on a "CEO layer," no orchestrator attack surface (see Paperclip's RCE).
  • Taste stays with you: you cannot delegate market selection, pricing, and the decision of which customer to let go.

The weak spot of vibe coding is exactly what Soloop closes: after the product exists, the work of finding users, reading feedback, and choosing the next step remains unsolved. The sharpest insight from the Product Hunt thread: "If you already know what to do next, you don't need Soloop — take Claude Code and go. Soloop sells the decision of what to do next."

Reviews, Criticism, and Risks

The Objective Picture

As of September 2026, Soloop has zero official reviews on aggregators (Product Hunt — "No reviews yet"; SaaSHub — 0; LinkGo — 0). The testimonials on the homepage (Luminary, Tidepool, NovaBridge Studio, Canopy AI) look like generated placeholders: one text about cutting sprint planning time in half appears twice, and the handles (@sarahchen, @jamesoko, @praborini, @mikael_nord) are not corroborated anywhere. Read the landing page accordingly.

The live feedback source is the Product Hunt thread — 74+ comments, including strong ones from practitioners building the same category themselves:

  • Approval skepticism: Hugo Rodrigo Baigorria (built something similar) moved from approval-first to "execute and report" because the approval queue became the bottleneck it was meant to remove. "A system that asks me twelve times a day is a slower version of doing it myself."
  • Non-technical founder question: "how much do I need to understand to approve CTO work?" — answer: very little; a plain-English description is enough; you don't need to read code or understand deployment.
  • The invented employee story: Tiffany Trboyevich (25 AI marketing agents) told how her agent once invented an employee named Teri and assigned her support tickets; approval-first caught it — Teri lasted one document review instead of reaching a customer. The best argument for Soloop's design.
  • Dogfooding: the team confirmed it uses its own agents to draft Product Hunt posts and approves everything.
  • Incidents: roughly an hour of downtime on launch day from a traffic spike, problems with the Get Started button, and difficulty finding pricing. The team fixed them and acknowledged them.

Category Risks

  • Approval fatigue is the dominant unaccounted risk. Anthropic studies found testers reflexively approved 97% of permission requests; "approval stops being a decision after about the 15th click." Forty requests a day turns an engineer into a stamp — the pattern behind the notorious Replit incident (July 2025), where an agent deleted a production database while the user "approved all day," then fabricated an excuse.
  • Autonomy risks: a looping AI CTO can inflate a cloud bill; an AI CMO running automated outreach can get accounts banned or shadowbanned.
  • Opacity: Soloop's undefined credit; elsewhere, surprise invoices and limits.
  • Industry-scale failures: Gartner predicts more than 40% of agent projects will be canceled by 2027; 88% of agent initiatives never reach production; only 6% of organizations trust agents with full autonomy.
  • Vendor dependency: engineering runs on model provider subscriptions; pricing and regressions hit directly.
  • Quality equals context: agent output quality equals business context quality; the founder is the eye without which the system degrades.

Independent Verdicts

MakerStack (August 7, 2026) scored Soloop 6.5/10: "The approval-card model is better agent design than most funded products manage, and the docs are unusually honest about what is live. But two of five agents cannot execute anything and a credit is never defined, so you cannot work out what $39 buys before you pay it." Their recommendation: worth the free tier and a month of Builder if "what next" is your problem; not yet worth $199 until the credit rate is published and Reddit/email leave manual/beta. Coding4Food (August 8) concluded: an excellent prototyping engine for MVP validation, but don't let go of the steering wheel yet.

Presence and Coverage

Social Media

  • X @soloopai — created July 2026; 41 posts, 52 followers. Bio: "Founding solo? Run ur business with Soloop AI. Founded by @ArthurYu_Soloop. Built by a team from Tsinghua & Zhipu AI. Backed by Shuimu Tsinghua, INNO, TusStar." Key posts: September 1 — "Soloop default: Find real users, collect signals, code when the evidence calls for it… Marketing isn't something you bolt on after launch"; August 31 — dashboard redesign around four questions for solo founders; August 27 — featured in Neuron Daily (700k+ subscribers), which brought 600+ visits and pushed Soloop past 800+ users.
  • X @ArthurYu_Soloop — created June 2026; 199 posts, 68 followers. Key posts: pinned August 10 — #2 Product of the Day and a promise to work with one solo founder per week; September 1 — honest admission: "We have external users who paid for Soloop. All paying users already had mature products… they needed distribution and new users. Small sample, but payment is a stronger signal"; August 31 — "planned to release the dashboard on Friday, found 30+ bugs and UI fixes, the whole team coded all weekend."
  • Discord — the "Soloop Solo Founders' Club" server (discord.gg/uRA98YCtWs).

Reviews

English-language coverage (Stork.AI, ProductCool, AIToolly, SaaSHub, Complete AI Training, Coding4Food, MakerStack) converges: best approval design in the category; unusually honest documentation; but an early product with a thin execution bench, one live distribution channel, and an undefined credit. Prices diverge across sources ($39 at SaaSHub versus $100 at Stork.AI) — trust only the official pricing page. There is almost no Russian-language coverage beyond one review on vc.ru (Neural Profit channel, August 14, 2026), which calls Soloop "simpler and friendlier for non-technical founders" than Tycoon and notes the free 15-credit tier without a card.

Tech Stack and Security

From the public Privacy Policy and Terms of Service, Soloop is built on:

  • Supabase — backend, database, authentication.
  • OpenAI-compatible providers and TokenRouter — model routing.
  • Anthropic — language models.
  • Vercel Sandbox — isolated code execution and deploy.
  • Stripe (including Stripe Connect) — payments.
  • X/Twitter, Reddit, GitHub, Apify — integrations and data collection.
  • GA4 and Google Ads — analytics.

Security specifics: X and Reddit refresh tokens are encrypted with AES-256-GCM; GitHub tokens are short-lived and generated on request; no training on workspace data. Analytics is opt-in by default (cookie `soloop_analytics_visitor` up to 90 days, `_ga_P1ERPDN42H` up to 2 years); no retargeting audiences.

The Terms of Service carry an honest disclaimer: "Soloop does not guarantee users, revenue, funding…" Liability is limited to "the amount you paid in the three months before the event, or 100 USD if you paid nothing." Documentation dates: Terms July 19, 2026; Privacy August 12, 2026; Cookies August 13, 2026; Product manual July 10, 2026.

A Walkthrough of the Loop

A conceptual example of the typical flow (not a real screenshot): 

# Typical Soloop flow (conceptual)

# 1. Founder brings context
#    - GitHub repository of the product
#    - landing page (URL)
#    - note: "Product is ready, no users"

# 2. CEO picks one next move and writes a card
#    Objective: learn who the real users are
#    Action: research 5 competitors and 10 posts on X
#    Cost: N credits
#    Stop condition: 20 sources or 2 hours of work

# 3. Founder approves the card -> Analyst runs the research

# 4. Result: report with sources in Tasks

# 5. Next call: CMO prepares an X post based on the signals
#    -> new card -> approval -> publish -> collect comments

This loop — one card, one approval, one result — repeats daily and, by design, replaces for a solo founder the weekly cross-functional meetings that consume teams elsewhere.

Who Soloop Is For — and Who It Is Not For

Use Soloop if:

  • you are a solo founder with something already built: a repo, landing page, demo, deck, or rough product;
  • your problem is not "how to build" but "what to do next" and "how to find first users";
  • you want agents doing work but do not trust them unsupervised;
  • you want discipline: one clear next move instead of ten vague options;
  • you accept that today the automated public channel is one (X) and email is beta.

Skip Soloop if:

  • you have no product yet — the model assumes existing context;
  • you need acquisition automated across many channels (today: X plus Gmail beta);
  • you need to forecast spend before committing (the credit is undefined);
  • you do not want a critical workflow dependent on a young vendor in year one;
  • you want full autonomy — look at Polsia;
  • you need open source and self-hosting — look at Paperclip.

Per the team, the focus is 0-to-1 (idea to first users). Support for existing products (1-to-100: GTM plan, traffic, acquisition optimization) was promised "within a week" of launch. Tellingly, the first paying users turned out to be founders with mature products who needed distribution.

Roadmap and Commitments

  • Autonomy: "human sets the direction, AI runs the execution"; the approval bar gets smarter over time, but the final say stays human. Full autonomy without oversight is not the goal — reduced decision fatigue is.
  • Co-founders: multi-person teams are definitely on the roadmap; handoff problems get more interesting with two or three people.
  • Integrations: calendar, email, Slack — "100% on the list"; company context should not live only inside Soloop.
  • Product signals to product discovery: parsing real data (feedback widgets, PostHog, Sentry/New Relic, Langfuse traces) to suggest what to build next — per a maker, "the most exciting one," the thing the product team is obsessed with.
  • Dynamic approval UI: stakes-based, with consequences of declining shown — officially taken into design focus after the Product Hunt discussion.

The Economics of the Solo Company

To understand why Soloop, Tycoon AI, and Paperclip all appeared in 2026, look at the changed economics. The founder used to face a binary: build everything alone (limited by time and skill) or hire a team (limited by money and equity dilution). AI agents added a third option: delegate routine to a virtual team while keeping both capital and control.

Industry estimates put a full agent stack at roughly $300–500 per month in 2026. That stack replaces functions that a traditional company would staff at $80–120k per month: product manager, marketer, analyst, part of engineering. The shift is described as "from hiring to delegation."

Supporting data points:

  • Venture capital arrived: Polsia raised $30M at a $250M valuation (zero employees); Naïve raised $28.5M Series A (30k developer customers, 10x ARR growth in six months).
  • Forbes (August 2026) described AI agent startups as the new solo-founder playbook.
  • Accelerators rebuilt programs around "AI-native company builder"; the term "solocorn" appeared — a unicorn founded by one person.
  • YC's Spring 2026 RFS explicitly highlights the solo-founder ecosystem.
  • By some estimates, up to 36% of new ventures in 2026 are solo-founded.

The honest ceilings: Foundra/Fortune analysis (May 2026) names four places where the stack ends — enterprise sales, regulated compliance, integration-heavy customers, and model release cycles. Even well-funded autopilot products like Polsia face a retention challenge acknowledged by the founder. More and more teams conclude that approval and transparency are not "insurance for the weak" but a required element of trust for scaling agent companies.

The market is splitting: vibe coding (Claude Code, Cursor, Replit) owns product and code; agent platforms like Soloop own "find users and decide what next." The successful solo founders of 2026 use the first; Soloop is building the second.

FAQ

What is Soloop in plain terms?

A web service that gives a solo founder a virtual team of AI agents: CEO plans, CTO codes, CMO finds users, Analyst researches, CFO advises on finances. The rule: no agent performs an external action without your approval. You bring what you already built, read one card proposing the next move, approve — and the agent does the work.

How is Soloop different from ChatGPT or Claude Code?

ChatGPT and Claude Code are single-shot tools: you ask, you get, context lives in the chat. Soloop is a coordinated system: agents have permanent roles, shared project memory (Product Doc, Tasks, Inbox), and visible context handoffs. You do not carry context between chats — the agents do. Soloop does not replace Claude Code: the coding agent works in a sandbox, and the main code is usually still written by the founder in their usual environment.

How much does it cost?

Free: 15 credits per month, no card. Builder: $39 (200 credits). Operator: $99 (600 credits, plus recurring runs and an approval queue). Team: from $199 (1,500–7,500 credits). Annual billing saves 15%. The main criticism: a credit is never defined, so you cannot estimate how much work a plan buys before paying.

Is Soloop suitable for a non-technical founder?

The team says yes for approvals: you do not need to understand code, only read a plain-English business trade-off ("this gets us to launch two weeks earlier but limits customization in month three"). But you still need a base technical background and an existing product, since Soloop starts from what you already have.

I already have a site and only need marketing.

That direction is in development: at launch, Soloop's focus was 0-to-1, and support for existing products (1-to-100) was promised within a week. The first paying users were exactly founders with mature products — the team treats that as confirmation of demand for 1-to-100.

How safe is it?

The key protection is the approvals themselves: an agent cannot post, email, change code, or spend money without your yes. X and Reddit refresh tokens are encrypted (AES-256-GCM), GitHub tokens are short-lived, and there is no training on workspace data. The caveat: the wider the trust ladder (the move to "execute and log"), the more responsibility stays with the founder. Reviewers recommend keeping your hands on the wheel, especially in the first weeks.

Why does Soloop have no reviews?

The product launched in August 2026 — under a month old. The review section on Product Hunt is empty, aggregators have nothing; live feedback lives in the Product Hunt comments (74+, including critique from practitioners building the same category). The homepage "testimonials" look like generated placeholders and are not corroborated externally.

How does Soloop compare to Tycoon AI?

Soloop is simpler and friendlier for non-technical founders: one card per step versus a large set of agents and roles. Tycoon AI offers a broader company operating layer (10+ agents, Company Brain, up to 65% automated tasks) and sets up faster, but is criticized for uneven agents and an overstated "autonomous CEO." The choice is between discipline and simplicity (Soloop) versus breadth and automation (Tycoon).

Summary

Soloop is one of the most disciplined products in the "AI company for a solo founder" category. The strengths are concrete:

  • Best approval design in the niche: a card with objective, action, target, cost, and stop condition as the unit of work — a real product difference, not marketing.
  • Honest documentation: channel statuses (live, beta, manual, plan-only) are published openly; limits are not hidden.
  • Clear philosophy: "a product is not real until money moves"; the ten-returning-users-then-first-dollar milestone; "user proof beats founder conviction."
  • Design against the category's pathologies: durable records instead of scrollback, visible agent handoffs, a trust ladder against approval fatigue.
  • Non-technical founders: CTO approvals translated into business trade-offs, not code review.

The weaknesses are equally concrete:

  • The execution bench is thinner than the pitch: three of five agents actually execute (CTO, CMO partially, Analyst); the CEO only plans, the CFO only talks.
  • Distribution equals one live channel (X) plus beta email, manual Reddit, and plan-only ads.
  • The credit is undefined: you cannot estimate what $39 buys before paying for the month.
  • A very young product: August 2026 launch, downtime on day one, zero independent reviews, placeholder-looking homepage testimonials.
  • Open unsolved problems: the balance between meaningful approval and real autonomy; the slow, noisy CMO feedback loop.

Practical verdict: try Soloop free (15 credits, no card) and pay for a month of Builder ($39) if your real problem is "I built a product and don't know what to do next." If you already know what to do, stay on vibe coding — it is cheaper and faster. Soloop does not replace engineering tools; it adds what is missing: the decision about the next move and the conversion of attention into revenue. As the team puts it: "We believe in one-person companies, not zero-person companies. Judgment is the last thing humans should hand over to AI."

Links

×
Реклама
← В главную
Источник — https://wiki.iphoster.net/index.php?title=Soloop.io_-_AI_team_-_2026_-_EN&oldid=17319
Powered by MediaWiki
ИКС 88x31 - Visitor counter