$17.8K MRR: How Lancer's AI agent finds and bids on Upwork jobs

What it sells and to whom

• Who pays: Upwork freelancers and agencies. The homepage says it's "Trusted by 400+ freelancers and agencies".
• Price: the homepage's structured data lists Pay-Per-Lead at $149/month ($299 per quarter) and Unlimited at $499/month ($999 per quarter). In a Mar 2026 interview the founder described earlier plans: $79 for 30 proposals, about $300 for 250, and $500/month unlimited as the launch offer.
• What they get: AI job filtering, "unlimited targeted campaigns" with filters on client history and spend, personalized cover letters and answers to screening questions written from the user's knowledge base, Discord/Slack alerts, analytics and onboarding videos.

How the agents run it (what's automated)

Founder Ivan Nedelkovski explained the reason on Starter Story. There are "around 200 000 jobs posted on [Upwork] each month" across all hours, and "it takes around 10 minutes to qualify a job and then write a personalized proposal". That made it "a perfect case where an agent would outperform a human."

LLMs do two jobs in Lancer, per Ivan: job qualification and proposal writing. The FAQ describes two layers. Filters can "evaluate jobs based on 99+ variables", then "our AI then reviews each job that passes the filters, just like a human would". The homepage says it bids "within 10 minutes of job being posted".

Users can stay in control with "manual approval mode" and set auto-apply delays. Once they trust it, Lancer recommends "letting Lancer do it's thing on auto-pilot."

What still needs a human

• Distribution. Lancer grew with "zero paid ads", using what Ivan calls "connectors": Upwork coaches who already have Lancer's buyers as clients. He pays them "30 [percent] commission for a lifetime" if they onboard customers, or 20% for a plain referral (Starter Story). Recruiting and managing those partners is people work. The first coach came through a beta user's intro. For the second, Ivan "straight up just offered him 1 000 to jump on a call".
• Platform risk. Lancer depends on one platform. Its FAQ notes "Upwork prohibits spammy bots that mass-apply to jobs" and says it stays within the terms. Agent Money List's entry also records that the team spent about 4 months on a shared "Agency Manager" account approach before rebuilding on direct account connections (Indie Hackers).
• Product upkeep. Ivan runs a dev shop. His TrustMRR message calls Lancer "a side-project that took off - no time for growth". So even this agent business has a team behind it: Indie Hackers reports "a two-person team".

Agent Money List rates Lancer nearly autonomous (our estimate). The agent delivers the service, while the team handles growth, support and edge cases.

Tools/stack

Sourced from the founder (Starter Story) and TrustMRR:

• App: TypeScript, Next.js and Node.js
• Hosting: Hetzner and GCP (Starter Story also mentions Firestore for the MVP)
• LLMs: accessed through OpenRouter
• Search: Elasticsearch for querying jobs
• Account access: "various proxy providers" for connecting safely to Upwork accounts
• Affiliates: Tolt
• Payments: Stripe
• Coding: "we're all using a cursor with Opus 4 5"

Numbers

• Verified MRR: $17,783 (TrustMRR via Stripe, last updated Oct 16, 2026; from entries.json). The same snapshot shows $16,039 for the last 30 days and $211,406 all-time.
• Latest page check (Oct 20, 2026): TrustMRR now shows MRR $17,979, 114 active subscriptions, $15,499 for the last 30 days and $212,068 all-time. It lists the business as founded April 2025, in Estonia.
• Founder claims (not verified by us): $10K MRR within two months of launch and "$20K MRR with a two-person team" (Indie Hackers, Jul 29, 2026); "now doing $27K months" (TrustMRR founder message). The verified figures above are what we rank on.
• Self-reported on the homepage: "2,884 Upwork client conversations started by Lancer users in the last 30 days"; "60% open rate. 21.8% reply rate. 12.5% win rate".
• Our estimates for a copycat: about $185 to start and about $25/month to run (entries.json, labelled est.).

How you could build a version of this hands-off

Ivan's own advice: "every platform which has a huge user base is something that you can automate", because "the users basically live on the platform" and you can find them.

1. Pick a platform where people do repetitive, rule-based work. Freelance marketplaces, resale sites and job boards are examples. Read the platform's terms first, because your whole business depends on them.
2. Size the real market. Count the active users who would pay, not the registered ones, and check what competitors already earn.
3. Build the two-layer brain. Use hard filters first (cheap), then an LLM review of the jobs that pass, then an LLM writes the output. Keep a manual-approval mode for new users.
4. Connect to the user's own account safely. Lancer's costliest mistake was a shortcut here. Use the user's own access and slow, human-like pacing.
5. Recruit "connectors" instead of running ads. Find coaches or influencers who already serve your buyers. Offer recurring commission and track it with an affiliate tool.
6. Price on output. Charge per proposal or lead for small users and a flat unlimited tier for agencies.