Last week I posted one line about the new Fable model and it would not stop moving. The line was simple: Fable is the first model that behaves like it is telling you, I know what I am doing, stay out of it please. It pulled 131,709 impressions and 784 reactions, and the comments turned into an argument about whether AI can really run work on its own yet. Under the noise sat the question I actually care about as someone who sells every week. If the model is finally confident enough to act without hand-holding, is the AI SDR everyone keeps selling us real now, or still a demo dressed up for a launch video?
I run Growth Cab, the GTM advisory I founded, and outbound is not a theory for me. I do it, my team does it, and we build the tooling we use. So this is not a review of a product I watched once. It is what I have learned handing real prospecting work to an AI SDR built on this new class of models, including the parts that still embarrass me.
in
“Fable is the first model to go like: Bro I know what I'm doing. Stay out of it please. That confidence makes sense when you see how Anthropic shipped it.”
What an AI SDR Actually Is Now
An AI SDR is software that does the job of a sales development rep. It picks the accounts worth chasing, researches each one, writes the first message, handles follow-up, and tries to book the meeting. The category has lived on slides for two years. For most of that time it was a mail merge with a confident marketing page, and anyone who ran real outbound could feel the gap inside a week. The messages were generic, the targeting was lazy, and the whole thing produced volume no serious buyer would ever answer.
What changed is not the software wrapper. It is the model underneath. The Fable-class models research like an analyst and make the small judgment calls that used to require a person. That single improvement is the reason the AI SDR went from a party trick to something I actually keep in the pipeline.
Why the Model Jump Is the Whole Story
Here is the workflow that convinced me. We built a skill on top of Claude that takes one target contact and does deep research on the company, reads how they describe their own product, finds the trigger event worth mentioning, and drafts a first touch that only makes sense for that one person. If you sent the same message to a different company it would fall apart, and that is the point. Relevance that specific was impossible when the model could not hold enough context or reason about it.
The older tools failed because they could not decide anything. They filled a template. A good SDR is mostly a sequence of tiny decisions: is this account worth the effort, what does this person actually care about, is now the right moment, what do I say that earns a reply. The new models get enough of those decisions right that the output stops reading like spam. That is the real unlock, and it is why a founder who wrote off this category last year should look again.
The Three Jobs I Hand the AI SDR
I do not let it run the whole motion. I give it the three jobs it does better than a tired human at 8am.
One, research. It reads the account, the person, the recent news, and the tech footprint faster than I can open the tabs, and it hands me context I would have skipped. Two, the first draft. It writes a first touch specific enough to send after a quick edit, which deletes the blank-page tax that quietly kills consistency. Three, follow-up memory. It tracks who went quiet, what was said, and when to nudge, so nothing slips through the cracks between calls.
Those three alone gave me back the two hours of manual prep that used to sit in front of every good outbound day. That is a real number, and it repeats every single morning.
Where the AI SDR Still Breaks
Now the honest part, because the launch videos will never tell you this. Three things still fall apart, and two of them are baked into the economics.
First, cost and limits. The reason the output is finally good is that it runs on the best available model, and the best model is expensive and rate limited on purpose. My own post joked about it: you can burn a weekly usage limit in a handful of prompts and then sit in timeout for days. Point a fully autonomous AI SDR at a large list and you meet that wall fast. Quality at scale costs real money, and anyone promising unlimited personalized volume for a flat low fee is quietly turning the model down to survive the math.
Second, judgment about when to stay silent. The model is good at deciding what to say and weak at deciding whether to say anything at all. It will write a confident, well-researched message to an account that any experienced rep would skip. Left alone it optimizes for activity, and activity against the wrong list is how you burn a domain and a reputation at the same time.
Third, the actual conversation. The AI SDR can start a thread. It cannot yet hold the messy human exchange that turns a reply into a booked call, read the hesitation behind a short answer, or know when to drop the pitch and just be useful. The moment a real person answers with something unexpected, a human needs to take the wheel.
How I Actually Run It
So the setup that works for me is not a robot replacing the team. It is the AI SDR doing research, drafting, and follow-up memory, while a human owns the target list and every live conversation. The model brings the confidence my old post joked about. I bring the judgment about where to point it. That division of labor books meetings I am proud to take, and it does it without the two hours of grunt work.
If you are evaluating one of these tools, test it against that line. Ask it to prospect ten accounts and read what it produces. If the messages could have been sent to anyone, the model underneath is too weak or the vendor turned it down to save cost. If each message clearly belongs to one company, you are looking at the real thing, and it is worth building around.
Your Move
The whole debate under that Fable post came down to one instinct: people want to know if they can finally trust AI to run the work. For prospecting the answer is a careful yes, as long as you keep the judgment and the conversations human. Start with research and drafting this week and measure the prep time you get back.
I break down one AI revenue play like this every day in The Revenue AI Brief, the daily newsletter for B2B operators, including the exact prompts behind our prospecting skill. Drop your work email below and tomorrow's edition lands in your inbox. And if you think I am wrong about where the AI SDR breaks, tell me on LinkedIn. I answer everything.
