AI agents for sales development

Intro — how this agent works

An AI SDR agent researches a prospect, decides they are worth contacting, writes something personalised, sends it across email and LinkedIn, handles the reply, and books a meeting on an account executive's calendar. Every part of that works in isolation. The category's problem was never capability — it was that running the whole loop autonomously breaks the channel it depends on.

The outbound loop, step by step

What actually went wrong

By April 2026, Bain Capital Ventures stated plainly that fully autonomous AI SDRs had not replaced human sales teams at any meaningful scale. Three failure modes explain most of it:

• Deliverability is a shared resource you can burn. Cold outbound routinely runs 0.5–1% complaint rates without aggressive list hygiene. An agent that can send ten times more email does not get ten times the pipeline — it gets its domain filtered, and the cost shifts from labour savings to domain and brand spend.
• Personalisation collapses toward templates at scale. The thousandth "I noticed you're hiring SDRs" is a template whether or not an LLM wrote it. Recipients pattern-match faster than models vary.
• The metric was meetings booked, not revenue. Reported head-to-head tests put human SDRs at 71% meeting show rates against 52% for AI. Booking a meeting nobody attends is a measurable success and a commercial nothing.

The churn numbers followed: 50–70% annually across the category by multiple reports, with 11x reported at 75% three-month churn. That is not a product tweak away from fixed — it is a signal the unit sold did not match the value delivered.

What replaced it: GTM engineering

The money moved one layer down. Instead of buying an autonomous agent, teams hired a "GTM engineer" — 400+ open roles at a roughly $160k median — and gave them a programmable data layer. Clay is the winner of that shift: roughly $150M ARR by May 2026 and a $5B valuation, selling the enrichment and orchestration substrate rather than the agent on top.

The working 2026 pattern is hybrid, and about 45% of sales teams are reported to run it: AI does the research, enrichment, list building and drafting — the roughly 70% of SDR time that is admin — and humans own the conversation. That is a real productivity gain. It is not the "fire your SDR team" pitch that sold the category.

The honest test before you buy

• Are you buying volume or precision? If your TAM is 2,000 accounts, more email is the wrong lever entirely and no AI SDR will help.
• Who owns deliverability? If the vendor's answer is "we rotate inboxes", ask what happens to your primary domain when complaint rates climb.
• What is the contract length against the reported churn? Annual prepay into a category with 50–70% churn is a specific risk, not a general one.
• Measure show rates and pipeline, never meetings booked. The gap between those two numbers is where this category hid for two years.

Part 1 — the providers

Three groups now: the autonomous AI SDRs that defined the category, the data and orchestration layer that beat them commercially, and the incumbent sales platforms absorbing both.

Published AI SDR pricing is unusually unstable — several vendors moved to quote-only during 2026 and third-party reports disagree. Treat every figure here as an anchor to verify, not a quote.

Part 2 — how each provider handles the work

Autonomous AI SDRs — the agent owns the loop

• 11x — the company that defined the category, selling Alice for outbound and Julian for voice. Published pricing has been inconsistent (roughly $3,750/mo billed annually, or a stated $36,000/year, with buyers reporting a median near $40k), which is itself informative. It runs the full loop autonomously, which is exactly the model that drew the churn reporting — 75% at three months in one account.
• Artisan — sells Ava as an "AI employee", bundling B2B data, enrichment and deliverability management so you are not assembling a stack. Pricing moved to quote-scoped during 2026, with tiers described by volume (~2,500 leads/month on Team, ~6,000 on Scale) rather than dollars; third-party reports range from $280 to $5,000 a month. Bundling data is the genuine differentiator — it removes the most common failure point, which is bad lists.
• AiSDR — the cheap entry at $250/month for a solo tier. Useful for testing whether outbound works for you at all before committing five figures, provided you accept it is a lighter product.

Hybrid — agents inside a human workflow

• Regie.ai — the clearest expression of the correction: it sells per seat to teams that still have reps. AI SEP runs $180/user/month on annual with a 10-seat minimum; the Force Multiplier Rep tier is $499/user/month with a 5-seat minimum; the full RegieOne platform is quote-only, with third parties reporting $2,000–6,000/month. Seat-based pricing signals the agent augments a rep rather than replacing one — which matches what actually works.
• Qualified (Piper) — attacked inbound rather than outbound, which sidesteps the entire deliverability problem: Piper engages visitors already on your website, where the intent is real and the channel cannot be spam-filtered. Salesforce closed its acquisition on 1 April 2026 and folded Piper into Agentforce. That inbound-first thesis is why it got acquired while outbound-first peers churned.
• Outreach and Salesloft — the incumbent sales engagement platforms adding agentic features to sequences reps already live in. Less exciting, much stickier, and they own the workflow the AI-natives had to persuade reps to leave.

The data layer — where the value actually accrued

• Clay — a programmable GTM environment: waterfall enrichment across dozens of data vendors, then LLM calls over the results to research, score and draft at row level. It sells the substrate rather than the autonomy, and it won — roughly $150M ARR by May 2026, up from $108M at the end of 2025, a $5B valuation after a January 2026 tender, and median contracts around $30k/year. It also requires someone to operate it, which is what "GTM engineer" means.
• Unify — combines intent signals with sequencing so outbound fires on a trigger rather than a schedule, at $1,740/month billed annually. Raised a $40M Series B led by Battery at a $260M valuation.
• Common Room — aggregated person-level signal across community, product and social surfaces; Zoom agreed to acquire it in July 2026, another data-layer exit rather than an agent exit.
• Apollo.io — database plus sequencing at self-serve prices. Not an agent, but for a large share of teams it is the honest alternative: the data and the sending, with a human writing.
• Build your own — Clay or a data API, an LLM for research and drafting, and your own inbox infrastructure. This is what most sophisticated GTM teams actually run in 2026, and it is why the autonomous vendors struggled to hold accounts: the pieces are assemblable by one competent operator.

Part 3 — pros and cons

How to pick, in 30 seconds

• Testing whether outbound works at all → AiSDR at $250/mo, or Apollo plus a human writing.
• You have reps and want leverage → Regie.ai. Per-seat augmentation is the model with evidence behind it.
• You have website traffic going unworked → Qualified/Piper. Inbound intent beats any outbound agent.
• You have or will hire a GTM engineer → Clay, and skip the autonomous layer entirely.
• You want one vendor accountable end to end → Artisan, because bundled data fixes the most common failure.
• You were about to prepay a year for an autonomous SDR → don't. Pilot quarterly and measure show rates and pipeline, not meetings booked.

The investor read

• The picks-and-shovels layer won outright. Clay at $5B and roughly $150M ARR, against a cohort of autonomous SDR vendors reporting 50–70% churn. When the application layer churns and the data layer compounds, the moat was never the agent.
• Churn exposed a unit-of-value mismatch. These tools were priced like an employee replacement (~$36–40k/year, the cost of a junior SDR) while delivering a productivity tool. Regie's per-seat pricing and Qualified's inbound focus are the two corrections that stuck.
• Deliverability is a commons, and the category over-grazed it. Any product whose output degrades as adoption rises has a structural ceiling. This is the clearest example in enterprise AI, and worth checking for in any "AI does outbound X" pitch.
• Inbound is the defensible half. Qualified got acquired by Salesforce; outbound-first peers did not. Where intent is genuine and the channel cannot be filtered, agents work — and that logic generalises well beyond sales.
• Consolidation is running through the data layer. Salesforce bought Qualified, Zoom bought Common Room. The acquirers are buying signal and workflow position, not autonomy.

Frequently asked questions

How much does an AI SDR cost?

Published pricing runs from $250/month for AiSDR's solo tier to about $3,750/month for 11x's Growth plan billed annually — roughly $36,000–40,000 a year, deliberately close to a junior SDR salary. Regie.ai charges per seat instead, $180–$499 per user per month with seat minimums. Several vendors moved to quote-only pricing during 2026.

Do AI SDRs actually work?

Not as autonomous replacements. Bain Capital Ventures noted in April 2026 that fully autonomous AI SDRs had not replaced human sales teams at meaningful scale, and reported head-to-head tests showed human SDRs generating 2.6× the revenue with 71% meeting show rates against 52%. What does work is the hybrid model — AI handles research, enrichment and drafting; humans own conversations.

Why do AI SDR tools churn so much?

Reported churn is 50–70% annually, roughly double human SDR turnover. Three causes: cold outbound deliverability degrades as volume rises, personalisation converges on templates at scale, and the tools were measured on meetings booked rather than pipeline or revenue — so the gap surfaced only at renewal.

Is Clay an AI SDR?

No, and that distinction explains its success. Clay is the data and orchestration layer underneath — waterfall enrichment plus LLM research at row level — that a human operator, the "GTM engineer", drives. It reached roughly $150M ARR and a $5B valuation while the autonomous agents above it churned.

What is a GTM engineer?

The role that emerged to operate tools like Clay: someone who builds go-to-market systems — enrichment waterfalls, scoring logic, triggered sequences — rather than sending emails manually. There are 400+ open roles at a roughly $160k median salary, which is the clearest evidence that outbound became an engineering problem rather than a headcount one.

Will an AI SDR hurt my domain reputation?

It can, and this is the risk to underwrite before buying. Cold outbound runs 0.5–1% complaint rates without aggressive list hygiene, and an agent that sends far more volume reaches filtering thresholds faster. Ask any vendor what happens to your primary sending domain when complaint rates climb — "we rotate inboxes" is not an answer.

Other AI agent verticals

Sources

Pricing and traction read in August 2026: AI SDR pricing index, Artisan tiers, 11x and Regie rates, Unify. Funding and outcomes: Clay revenue and valuation, Clay funding history. Category performance: why first-generation AI SDRs failed, AI vs human SDR comparison. Churn, show-rate and head-to-head revenue figures are reported by vendors and analysts with commercial interests in the answer — directionally consistent across sources, but not independently audited.