Hlib Storchak · 2026-07-18 · ~10 min readQuick answer
AI sales tools split into three categories this year: plug-and-play AI SDRs that run a full motion end to end, enrichment and signal agents that do one data job well, and custom-built agents you assemble yourself on a coding-agent framework like Claude Code. Pick based on how much of your process is already proven, not on which demo looked the most autonomous.
Why the category matters more than the vendor
Every week I get asked which AI SDR tool to buy. That is almost always the wrong first question. Before "which vendor" comes "which category," because the three categories the market has settled into in 2026 solve different problems, and picking the wrong one means you are comparing tools that were never competing with each other in the first place.
I have sat through demos of an all-in-one AI SDR platform and a narrow enrichment agent in the same week, for the same client, and watched the client try to score them on the same rubric. They cannot be scored the same way. One replaces a chunk of your outbound motion. The other makes an existing motion better. Confusing the two is how teams end up buying a Ferrari to deliver groceries.
The three categories, at a glance
Here is the split as it actually exists in the market right now, not as any single vendor's positioning describes it.
| Category | What it does | Who it replaces or augments | Where it breaks |
|---|---|---|---|
| Plug-and-play AI SDR | Runs research, sequencing, sending, and reply handling as one packaged motion | Replaces a chunk of SDR headcount | Nuanced objections, senior buyers, anything outside the trained script |
| Enrichment / signal agent | Finds contacts, scores accounts, or surfaces buying signal, nothing else | Augments reps and existing tools | Does not send or book anything by itself |
| Custom-built agent | You assemble the workflow yourself on a coding-agent framework | Replaces point tools with something purpose-built | Needs someone who can build and maintain it |
Tip. If you cannot describe your current outbound process in one paragraph without saying "it depends," you are not ready for category one. Start in category two.
Category 1: plug-and-play AI SDRs
This is the category most people mean when they say "AI SDR." A platform like Artisan, AiSDR, or 11x takes on research, personalization, sending, and first-line reply handling as a single packaged motion, usually with a human reviewing exceptions rather than every message. The pitch is speed: you are live in weeks, not months, and the platform has already made the workflow decisions for you.
The tradeoff is the same one you take with any packaged product. You get less control over exactly how each step runs, and you are betting that the platform's trained behavior generalizes to your ICP and your voice. For a team with a proven, repeatable outbound motion and a clear reason to scale it, this category is often the fastest path. For a team that has never run outbound at volume before, it can hide problems in the underlying offer or list quality behind a layer of automation that looks like it is working.
Category 2: enrichment and signal agents
This category does one job and stops. It finds emails, scores an account against your ICP, flags a hiring or funding signal, or summarizes a call. It does not send anything and it does not touch your calendar. Tools like Clay, Explorium's enrichment plugins, and most waterfall-based contact finders live here.
The appeal is a low blast radius. A bad output from this category is a wasted look, not a burned domain or a prospect who got the wrong message. That makes it the right starting point for a team that has not yet proven its process is solid, and it is also the category that most Claude Code and Cursor based GTM tooling starts with before anyone lets an agent write anything.
Category 3: custom-built agents on a coding-agent framework
The third category did not really exist as a mainstream option two years ago. Now, wiring Claude Code or a similar coding agent directly into a CRM, an enrichment API, and a sending tool to build exactly the workflow you need is a realistic option for a GTM team with any technical capacity at all. You are not choosing between two vendors' opinions of how outbound should run. You are encoding your own.
This is also where the market's current general-purpose-agent race is most relevant. Anthropic expanded Claude Cowork from desktop to mobile and web on July 7, 2026, and The Information reported on July 9 that Cursor is separately building its own general-purpose agent, internally codenamed Sand, aimed at the same kind of everyday work Cowork targets. Neither of those tools is a GTM product. But both point at the same direction category three is heading: fewer purpose-built demos, more general agents that a GTM engineer points at a specific workflow. The catch is that Cursor has not committed to shipping Sand publicly, and its pending $60B acquisition by SpaceX could rewrite that roadmap before it does, which is its own lesson about not betting a custom build entirely on one vendor's product plans.
The market data behind the split
The category split is not just a framing exercise, the money is actually moving that way. Fortune Business Insights projects the AI SDR market growing from $4.12 billion in 2025 to $15.01 billion by 2030, a 29.5% compound annual growth rate. That is the headline number vendors quote. The detail that matters more for this framework is inside the same report: the API and add-on segment, the closest proxy to category two and three combined, is the fastest-growing slice at a 24.98% CAGR, even though full software platforms still hold the larger share today.
Read plainly, that says buyers are not simply picking the biggest all-in-one platform and calling it done. A meaningful and growing share of the market is choosing to plug narrower capabilities into their own stack instead of buying one vendor's full opinion of the outbound motion.
What GTM leaders actually trust, and where they don't
A GTM operator roundtable run by Chili Piper puts a sharper point on the same trend. The consensus that came out of it, as Chili Piper summarized it, is that leaders trust AI that does one job extremely well and do not trust AI that claims to run the entire sales process. Inbound AI agents, answering a website question instantly or qualifying an inbound trial request, get consistently good marks because the problem they solve is clear. Outbound AI results were described in the same roundtable as inconsistent at best, with one framing that stuck with me: inbound AI solved a clear problem, outbound AI is still searching for one.
That is not an argument against category one outright. It is a reason to be honest with yourself about which problem you are actually trying to solve before you buy a tool that claims to solve all of them at once.
A decision framework: which category fits your team
I use three questions with clients before any vendor conversation happens.
| Question | If yes | If no |
|---|---|---|
| Is your outbound process already proven and repeatable by hand? | Category 1 can scale it | Start in category 2 first |
| Do you need one narrow job done well, not a full motion? | Category 2 is the right size | Look at category 1 or 3 |
| Do you have someone who can build and own an agent long term? | Category 3 gives you the most control | Buy category 1 or 2 instead |
Most teams answer "no" to the first question more often than they admit. That is the honest reason category two absorbs so much of the market's growth even while category one gets the bigger demos.
Where teams get the category choice wrong
The most common mistake is buying category one to fix a category two problem. A team with a messy, unproven list and no clear ICP buys a full AI SDR platform hoping the automation will paper over the process gap. It usually does the opposite: it runs the bad process faster and with more confidence, and the team blames the tool a quarter later instead of the list they fed it.
The second most common mistake runs the other way: a team with a genuinely proven, high-volume motion stays stuck manually stitching together enrichment and sending tools long after a packaged category one platform would have paid for itself in saved hours. Both mistakes come from skipping the category question and going straight to a vendor bake-off.
Key takeaways
- AI sales tools split into three categories: plug-and-play AI SDRs, enrichment and signal agents, and custom-built agents on a coding-agent framework.
- The API and add-on segment, closest to categories two and three, is growing faster (24.98% CAGR) than the market overall (29.5% CAGR to $15.01B by 2030), per Fortune Business Insights.
- GTM leaders trust AI that does one job well over AI that claims the whole sales process, per Chili Piper's operator roundtable.
- Pick your category based on whether your process is already proven, not on which demo looks the most autonomous.
- The two most common mistakes are buying category one to fix a category two problem, and staying stuck manually stitching tools long after category one would have paid for itself.
Where the Forge stack fits in this
My own stack sits mostly in category two and three. Leadsforge handles enrichment and list data, squarely category two, and Salesforge plus Agent Frank handle sequencing and AI-led sending on top of a process I already proved out by hand first. I did not start there. I started with a manual motion, added enrichment tooling once the process was solid, and only automated sending once the override rate on drafts was low and stable. That ordering matters more than which category you eventually land in.
My take: which category I reach for, and when
For a client with no proven motion yet, I reach for category two first, almost every time. It is the lowest-risk way to learn what your actual process should look like before you hand any of it to something that sends on your behalf. For a client with volume and a repeatable motion, category one earns its keep, provided the team goes in clear-eyed about where it breaks: nuanced objections, senior buyers, anything outside the trained script. Category three I reserve for teams that already have a coding agent workflow running somewhere else in the business and a person willing to own it, because the control it gives you is only worth the maintenance cost if someone is actually going to do the maintaining.
FAQ
What are the three categories of AI sales tools in 2026?
Plug-and-play AI SDRs that run a full outbound motion end to end, enrichment and signal agents that do one data job well, and custom-built agents assembled on a coding-agent framework like Claude Code.
Which category should a team with no proven outbound process start with?
Category two, enrichment and signal agents. It has the lowest cost of a wrong output and helps a team learn what its process should actually look like before automating anything that sends or books on its own.
Is the AI SDR market really shifting toward point solutions?
The data points that way. Fortune Business Insights has the API and add-on segment, the closest proxy to point solutions, growing at 24.98% CAGR, faster than the 29.5% CAGR projected for the AI SDR market overall.
Do GTM leaders trust full-process AI SDR platforms?
Mixed. A Chili Piper operator roundtable found leaders trust AI that does one job well more than AI that claims to run the whole sales process, and rated outbound AI results as inconsistent compared to inbound.
Can I mix categories instead of picking just one?
Most working stacks do. A common pattern is an enrichment agent feeding data into a packaged AI SDR platform, or a custom-built agent that calls a point-solution enrichment tool as one step in a larger workflow.