There is a stretch in a company's life where the founder is also the SDR. It usually starts the day you realise inbound is not going to save you, and it ends the day you can afford a dedicated person to run outbound. For most companies that stretch is months. For some it is years. For me it has been the entire pre-revenue life of Coil.
This post is about that stretch. What it actually feels like, what the work decomposes into, and what changes about it now that you can hand large parts of the work to a capable agent. I will not pretend any of this is easy. I will not promise a flywheel. The point is to write down honestly what running your own outreach looks like in 2026, because the gap between the pitch and the practice is wider than most tooling vendors will admit.
Why the founder ends up doing it
There is a familiar arc. You raise a small round or you bootstrap. You ship something. You wait for the demand you assumed was already there. The demand is not there in the volumes you need, so you start writing outbound emails on a Sunday afternoon and never stop.
This is rational. Cold outreach is the only acquisition channel that does not require you to already have an audience, a brand, a content engine, or a paid budget. It works with a list, a thesis, and your own time. For a pre-Series A founder, those three things are the resources you actually have. So the math closes and the founder picks up the work.
What founders forget when they start is that outbound is four jobs, not one. The handbook version of cold outreach for founders covers them in depth. Quickly:
- List. Who are you reaching out to? On what criteria?
- Write. What is the first email? What is the rest of the sequence?
- Send. How do you get the message into the inbox, at the right time, without getting filtered?
- Learn. What signal comes back? Which assumption did each batch test, and what changed?
You can be good at one of these and still get nowhere. The list can be perfect and the writing can be flat. The writing can be sharp and the list can be wrong. The send can be technically clean and you can still be sending to people who are not in market. The learn step is the one founders skip most often, because by the time the results come in, they are already in the next sprint and the discipline of reading the batch carefully has slipped.
What each job actually costs the founder
The honest accounting is in hours per week, not dollars.
List building, done properly, is a half day per cohort. You decide a hypothesis (founders at seed-stage B2B SaaS, 5 to 20 employees, hiring an SDR right now, in North America). You translate that hypothesis into filters across whatever data source you use. You scan the first hundred rows and you find that twenty of them do not match. You tighten the filters. You enrich. You verify. By the time you have a clean cohort of two hundred prospects you trust, half a day is gone.
Writing the sequence takes another half day if you take it seriously. The first email is two or three paragraphs, but the four drafts before that are a paragraph each, then a half page each, then back down to two paragraphs after you cut the parts that sound like a brochure. Follow-ups are easier because the constraints are tighter (one idea per message), but the variants for different ICP segments multiply the work.
Sending in 2026 is mostly a setup problem. Warmed inboxes, domain rotation, throttles, suppression lists. Once it works, it works. But the first time you set it up, a founder can lose a week. After that, the cost is mostly maintenance and watching for the deliverability cliff that always eventually comes.
Learning is the cheapest job in clock time and the most expensive in attention. Reading thirty replies and noticing the pattern (three different people pushed back on the second sentence, in similar words) takes thirty minutes. The expensive part is then translating that signal into a change to the sequence and to the next list. Founders skip this because by the time the data is in, the next cohort is already sending.
Add it up. Twenty to thirty hours a week if you are doing all four jobs at a serious level. That is a part-time hire's worth of founder time, and it is the part of your week that is hardest to delegate because the strategy lives inside it.
What changes when an agent can do each job
This is the part of the conversation that has changed in the last year, and the part that vendors are wildly overselling. So let me be specific.
An agent cannot replace the founder on outreach. It can replace some of the typing.
The honest framing is: each of the four jobs has a part that is judgment and a part that is mechanics. The mechanics part is what an agent should be doing. The judgment part is still you.
For list building, the judgment is the hypothesis (who, why, when). The mechanics are the filtering, the enrichment calls, the dedupe against people you already mailed, the verification of role and seniority. An agent that can call your data source, pull a cohort matching your filters, deduplicate against your existing leads, and verify emails is doing the mechanics. That is the thing Coil's Prospect Scraper exposes to agents the same way it exposes itself to humans, behind the same API.
For writing, the judgment is the voice and the angle. The mechanics are the personalisation lines, the subject line variants, the follow-up cadence boilerplate. An agent can produce drafts faster than you can. The trap is that an agent will also produce drafts that sound exactly like every other agent's drafts, and your reply rate will reflect that. The discipline is to keep the agent in the role of researcher and first-drafter and to hold the final edit yourself.
For sending, the agent has the least to add because deliverability is mostly an infrastructure problem and a behavioural problem. What the agent can do is monitor: deliverability rate by domain, bounce-rate drift, open-rate decay across the warm-up period. The kind of work an SDR ops person would do, in plain sight, in your CLI.
For learning, an agent earns its keep. The signal in a reply set is not hidden, but it requires reading thirty messages with the same intent. An agent that summarises a batch back to you in the language of "three out of thirty mentioned X, this is up from one out of thirty in the previous batch" is doing work the founder will otherwise skip.
What this looks like in a real week
A pre-Series A founder who is running their own outreach on the agent-driven stack typically has a Monday that goes like this.
You spend an hour reviewing last week's batch. The agent has already produced the summary. You read it. You decide whether the hypothesis held, partially held, or did not hold.
You spend two hours rewriting the sequence based on that read. The agent drafts three variants of the new opener. You pick one, edit it, and write the rest of the sequence around it yourself.
You spend half an hour pointing the agent at the next cohort: a new hypothesis, a new set of filters, a target size. The agent builds the cohort, verifies, and stages it for review. You scan the first twenty and approve or send it back.
You spend the rest of the week on product, on calls with the prospects who replied, and on whatever investor or operational fire is burning that week. The agent runs the send and reports back.
That is the realistic version. Not a sales pitch and not a magic trick. It is the founder doing the founder-shaped parts of the work, and the agent doing the agent-shaped parts. The total founder hours come down from something like twenty-five a week to something like six. The quality of the cohorts goes up because the agent does the enrichment and verification you would have skipped at hour eighteen of the week. The reply rate is roughly the same, because reply rate is mostly determined by the underlying hypothesis and the underlying writing, both of which are still you.
A note on what does not work
A few things I would caution against, having tried them.
Do not let the agent write the strategic email by itself. The opener and the close of your sequence are load-bearing. Have the agent suggest, but keep the pen.
Do not skip the read on every batch. The temptation to fire and forget is huge. The whole point of running this stack is that the founder stays in the loop on what the market is telling you. If the agent is the only one reading the replies, you are not running outbound, you are running spam.
Do not optimise for volume before you have product/market fit signal on the cohort. Five hundred replies that say "interesting, send me more info" is not signal. Three replies that say "we have this exact problem and we tried these three tools" is signal.
Where to go from here
If you are doing your own outreach right now, the practical next step is to read the founder cold outreach guide for the longer treatment, look at what the Prospect Scraper actually exposes, and check the pricing so you know whether the agent runs are realistic for your budget at your current volume. If you want context on why Coil is built this way, the about page covers it.
This stretch of company life is hard. The agents do not make it easy. They do make it possible to do the work at a quality level the founder would otherwise have to either pay for or skip. That is the trade.