Lead-Gen Agency vs Outbound Software (2026): the Belkins Question, Refereed

Where agencies genuinely win

1. True zero-bandwidth situations - a two-person founder team mid-fundraise cannot operate anything, however light.

2. Skill transplant - a good agency's first 90 days include ICP sharpening and message-market testing you couldn't do alone. (This is the real product; the meetings are the receipt.)

3. Email infrastructure motions at volume - domain warming, deliverability ops, and list hygiene are legitimately outsourceable specialist labor.

4. Defined campaigns with end dates - a launch, an event push - where standing up internal capability makes no sense.

Where software wins (and why the trend line points here)

1. Economics, brutally. A $3,500/mo retainer delivering 10 meetings is $350/meeting. The published software-motion equivalents: Gallea AI's 14 meetings in 15 days, Klaar's 8 meetings/2 months at $149/mo - an order of magnitude apart. (The receipts.)

2. The voice problem. Agency outreach is written by someone renting your name. In warm, signal-based outbound - where messages reference your posts and your profile visits - authenticity isn't cosmetic; the motion physically requires your voice, which is why agencies default to cold lists (the only motion they can run without being you). Cold economics follow: 1 - 3% replies vs the warm band's 15 - 45%.

3. The learnings stay. Six months with an agency ends with meetings had; six months operating a system ends with a tuned ICP, a trained voice model, and source-level knowledge of what converts - assets that survive the vendor.

4. The signal layer can't be outsourced. Your profile viewers and post engagers respond to you. An agency can't harvest an orbit it doesn't orbit in.

The questions to ask any agency before signing (the diligence card)

If the verdict table points you agency-ward, six questions that separate the good ones from the retainer mills: (1) "Whose lists, and where from?" - purchased-list answers predict the cold-economics plateau; agencies that build from your ICP and signals are the minority worth paying. (2) "Who writes as us, and what's the review loop?" - you want named writers and per-message visibility, not monthly sample decks. (3) "What happens to the data when we part?" - lists, learnings, and sequence performance should contractually transfer; if the answer is vague, the lock-in is the product. (4) "What's your sending architecture on LinkedIn?" - an agency running extension or cookie tools on your accounts is carrying your restriction risk for free; demand the architecture answer in writing. (5) "Show me cost-per-qualified-meeting for three current clients in our segment" - not aggregate case studies; comparable, current, segment-matched. (6) "What would make you tell us to stop paying you?" - the honest ones have an answer (usually: when your in-house motion outperforms us), and the ones who don't just told you the engagement's design goal. An agency that passes all six is rare and probably worth it; the number that fail question one alone explains most of this page's search volume.

The email-era note (August 2026): one classic agency-justifying job shrank this year - "we handle the email side" meant infrastructure (domains, warmup, deliverability ops) that quality motions no longer need: Valley's sequences now send email natively from your own OAuth inbox (~30/day/seat, verified in-sequence, included in every plan). Volume email operations remain legitimately outsourceable; the follow-through email rail no longer is a reason to pay a retainer.