Synthesis
Selling agent services
Earning money by deploying agents for other businesses: agencies, forward-deployed engineers, roll-ups and niche products.
Three episodes, all from people with something to sell. Vas (who leads a firm placing forward-deployed engineers) describes the job as end-to-end deployment, advises starting with one workflow for a small first client, selling the outcome rather than the technology, and structuring work as audit, build and prove. His large figures (pay as a share of value, roll-up exit prices, 50,000 engineers) are hypothetical. Joel Blackman runs a restaurant-focused agency he says makes about 100,000 pounds profit a year and is automating it with three agents; his 50,000-a-month figure is a target, and current revenue is about 10,000 a month. Billy Howell and host Greg Isenberg float directories, newsletters and a Shopify experiment as small agent-run businesses, with monthly revenue figures that are the host’s claims, unsupported by examples. Useful across sources: begin narrow, own the whole outcome, and measure. Treat every income number as a pitch.
Key points
- Start with one workflow for a small first client and own it end to end — 1 episode. (Vas · Vas (Oct 2026) ▶ 47:09)
- Sell the outcome each buyer cares about, not the technology — 1 episode. (Vas · Vas (Oct 2026) ▶ 26:48)
- Use an audit-build-prove cadence with before-and-after checks — 1 episode. (Vas · Vas (Oct 2026) ▶ 51:32)
- An agent-run agency can scale with few staff, but figures are targets — 1 episode. (presenter · How Grok Bot Is Going To Make (Oct 2026) ▶ 00:35)
- Small agent-run products (newsletter, directory, store) are floated as businesses — 1 episode. (Billy Howell · Billy Howell (Aug 2026) ▶ 41:12) (presenter · Billy Howell (Aug 2026) ▶ 39:43)
Frameworks
- Skills of a top forward deployed engineer — Three skill areas plus communication. (Vas · Vas (Oct 2026) ▶ 40:30)
- Understand how the work really gets done (the systems of record and the function)
- Ship production code that calls into those systems with auditability, governance and security
- Know the AI layer: which model to use, what to trust it with, evals, harness, rollbacks
- Communicate with senior leadership and convince them
- Process-first deployment sequence — A four-week audit, four weeks of building, then a before-and-after check at three and six months. (Vas · Vas (Oct 2026) ▶ 51:32)
- Weeks 1-4: audit systems and build the process maps
- Weeks 5-8: build the agents
- 3 and 6 months later: show what it used to be versus now
- Five-day starter plan — A practice routine: process-map your own life, then offer one simple workflow to a small business. (Vas · Vas (Oct 2026) ▶ 45:41)
- Step 1: list every app that holds your stuff; note which wins, which disagrees, how to route (day not stated)
- Step 2: take 20 things you did last week (day not stated)
- Wednesday: write one process down step by step, with exceptions
- Thursday: sort every step into the four buckets
- Friday: reach out to small businesses and offer to do one workflow for them
- Niche newsletter plus niche directory flywheel — Pair a niche newsletter with a niche directory on a linked domain: the directory monetises through referral links, and warm referral leads can become newsletter sponsors. (Billy Howell · Billy Howell (Aug 2026) ▶ 41:12)
- Niche directory curated by a research agent and published by a coding agent
- Newsletter on the same niche
- Referral links from the directory for revenue
- Referral leads converted into newsletter sponsors
Tactics
- Take a small first client and start with one workflow — Pick a small business, start with one simple workflow, do everything for them, give a timeline, do it for free if you must, and charge more next time. (Vas · Vas (Oct 2026) ▶ 47:09)
Stories & examples
- PE firm with 26 portfolio companies: Rolling out AI across 26 companies, each with 10 departments and 10 workflows each, would need around 50,000 FDEs over three years. Lesson: Group the portfolio by shared system of record instead of tackling each company separately. (Vas · Vas (Oct 2026) ▶ 22:09)
- Shopify product research done unprompted: Asked to find a product like one he liked, the agent went to ImportYeti on its own to see what other stores source and their margins, then wrote a report per idea with a recommendation. Lesson: Agents can pick sensible research tools unprompted; ask for recommendations. (Billy Howell · Billy Howell (Aug 2026) ▶ 34:57)
Numbers
- Value an FDE could drive for a company (host’s examples: 5, 10, 25): 25 million dollars (upper end of 5/10/25 range) (host’s framing, approx.) (presenter · Vas (Oct 2026) ▶ 00:10)
- Hypothetical pay: 10% of $10M value delivered: 1 million dollars (hypothetical, no period stated) (host’s framing, approx.) (presenter · Vas (Oct 2026) ▶ 44:38)
- Exit price in the roll-up example (buy at 1 billion): 8 billion dollars (upper of 2/4/8, hypothetical) (self-reported, approx.) (Vas · Vas (Oct 2026) ▶ 07:14)
- FDE share of carry in PE example (buy at 1 billion, sell at 5 billion; ASR ‘.5%’): 0.5 % of value delta (garbled in ASR) (self-reported, approx.) (Vas · Vas (Oct 2026) ▶ 45:07)
- Fee for a later engagement: 6 figures (five or six figures, currency unstated) (self-reported, approx., currency unstated) (Vas · Vas (Oct 2026) ▶ 47:26)
- Agency profit per year: 100000 GBP/year (host’s framing) (presenter · How Grok Bot Is Going To Make (Oct 2026) ▶ 00:35)
- Target monthly revenue (agency): 50000 GBP/month (host’s framing) (presenter · How Grok Bot Is Going To Make (Oct 2026) ▶ 00:47)
- Current monthly agency revenue: 10000 GBP/month (host’s framing, approx.) (presenter · How Grok Bot Is Going To Make (Oct 2026) ▶ 05:44)
- Current client capacity before hiring: 10 more clients (5 to 10) (host’s framing, approx.) (presenter · How Grok Bot Is Going To Make (Oct 2026) ▶ 05:48)
Caveats & warnings
- The £50,000 a month figure is a target, not a result; he currently reports about £10,000 a month. (presenter · How Grok Bot Is Going To Make (Oct 2026) ▶ 05:44)
- The Shopify store is an unproven experiment, and Billy says he is not an e-commerce expert. (Billy Howell · Billy Howell (Aug 2026) ▶ 36:22)
- Agent-deploying engineers who drive $5M-$25M of value can be paid a slice of it. (presenter · Vas (Oct 2026) ▶ 00:13)
- Agent teams can do 2, 5, 10 or 25 thousand dollars a month (host’s cold-open claim; no example revenue is given). (presenter · Billy Howell (Aug 2026) ▶ 00:24)
How much to trust this
- Evidence: 3 episodes, 3 independent interviews (each video counts once).