Actively AI announced on April 28, 2026 that it has raised $45 million in Series B funding at a $250 million valuation, co-led by TCV and First Harmonic, with participation from Bain Capital Ventures, First Round Capital, and Alkeon. The pitch behind the round is a direct shot at the data model that has anchored enterprise sales for 27 years: instead of asking humans to push work through a CRM, Actively assigns a persistent AI agent to every account and lets the system work the pipeline continuously. Total funding now stands at roughly $67.5 million.
The customer numbers are not vendor PowerPoint. Ramp, valued at $32 billion in March, attributes tens of millions in new revenue to Actively's system, with AI-driven deals closing approximately 23% more often than traditional ones. Security platform Verkada says the platform doubled sales productivity, with reps booking around 25 meetings per month. Samsara has deployed Actively across a go-to-market team of more than 1,000 people spanning sales, account development, revenue operations, and customer success. Other named customers include Attentive and Ironclad. None of these are pilot logos. They are large, AI-native companies running real revenue through the system.
The framing matters more than the round size. The founders—Stanford AI researchers Mihir Garimella, 26, and Anshul Gupta, 27, who founded the company in 2022—are not selling a Salesforce add-on. They are arguing that Salesforce itself is the horseless carriage. "Salesforce's data model was built in 1999—literally the year Mihir and I were born," Gupta told Forbes. "It was actually the perfect data model for a world in which everything was done by humans…but the rug was pulled out from under them."
For enterprise buyers, the question is not whether one Series B startup will displace Salesforce. It is whether the architectural assumption underneath their entire revenue stack—that the CRM is the system of record and humans push the work through it—still holds in a world where AI agents are the workers.
What Actively Actually Does
Most "AI for sales" tools fall into two categories: enrichment layers (better contact data, better intent signals) and content generators (better email drafts, better call summaries). Both still assume a human rep is the unit of throughput. Actively rejects that frame.
Its platform creates a custom AI agent per account, built on top of foundation models and trained on the company's historical data. The agent does account research, drafts outreach, builds presentations, flags missed signals, and proposes next steps. It runs continuously, not in response to a human sitting down at a CRM. The internal language is "Intelligence-Led Revenue"—accounts are worked whether or not anyone is paying attention to them.
The infrastructure logic is what makes it interesting:
- Persistent context per account. The agent remembers every prior interaction, every signal, every conversation, without a human having to log it. The CRM becomes one of many inputs, not the source of truth.
- Account-level intelligence that compounds. Samsara's COO of GTM specifically said the use-case-by-use-case approach to AI in sales doesn't scale; what works is intelligence that lives at the account level and compounds over time.
- Integration over replacement. Actively plugs into email, Slack, and Salesforce itself. Customers don't rip out their CRM. They just rely on it less as the locus of decision-making.
That last point is the strategic move. By integrating with Salesforce instead of fighting it head-on, Actively gets to displace Salesforce as the system of intelligence while letting Salesforce remain a system of record. Over time, if the intelligence layer becomes where decisions are made and execution happens, the system of record becomes a database—and databases are commodities.
The Salesforce Picture
Salesforce stock is down approximately 30% year-to-date in 2026, but Salesforce is not standing still. Its Agentforce suite has generated more than $9 billion in sales for customers, $800 million in annual recurring revenue as of February's earnings, and is in use at over 23,000 companies. The company is expected to launch Agent Albert later this year, designed to analyze user data and take actions autonomously. CEO Marc Benioff has explicitly dismissed the "SaaS apocalypse" thesis and argues AI strengthens Salesforce's franchise rather than threatening it.
The honest read on Agentforce is that it is the largest, most enterprise-distributed agentic CRM in production, and it is being adopted at real scale. The honest read on the criticisms is also real: customers have reported (per The Information and the Wall Street Journal) that Agentforce produces wrong answers, struggles with nuanced questions, and has trouble pulling data from outside Salesforce. Salesforce disputes those characterizations. Both things can be true at once. Agentforce is winning some customers and losing others on quality.
The broader market signal is harder to dismiss. Figma is down more than 50% as Anthropic's Claude Design eats into its territory. Intuit and ServiceNow are each down roughly 40% on AI competitive concerns. The pattern is consistent: incumbent SaaS valuations are being repriced for a world where AI doesn't just augment workers using the software, but increasingly replaces the software itself with agents that operate against shared data.
This is what Actively is betting on. The company isn't predicting Salesforce will collapse. It's predicting that as more value moves from "where the data sits" to "what's intelligently done with the data," the relative weight of CRM-as-hub vs. agent-as-hub will keep shifting.
Why Sales Was The Last Function To Get This Treatment
Anshul Gupta's framing is the cleanest articulation of the opportunity: "AI is transforming every single business function. We now have Cursor and Claude Code for coding, Decagon and Sierra for support, and Harvey and Legora for legal. But sales, the most expensive function in most companies—which spend 30-40% of every dollar they're bringing in on GTM efforts—has not seen the same impact."
The reason sales lagged isn't mysterious. Coding has well-defined inputs (existing code, specs) and outputs (new code, tests). Support has well-defined inputs (a question) and outputs (an answer). Legal has well-defined inputs (a document, a question of law) and outputs (an answer or draft). Sales is messier: signals are diffuse, timing is non-deterministic, the relationship layer matters, and outcomes depend on stitching dozens of small actions together over weeks. That complexity is what kept sales tech focused on workflow tools rather than autonomy.
What changed is that frontier models can now do multi-step, long-horizon, account-specific reasoning at a level where the persistent-agent-per-account model becomes economically viable. Two years ago, "give every account a personal AI" would have been a slide. Today, with Ramp and Samsara reporting real revenue impact, it's a deployed system.
The Developer Decoder
If you are a platform engineer, RevOps lead, or technical founder evaluating where this matters in your own stack:
The "agent per entity" pattern is generalizing. Actively's per-account agent is the same architectural shape as Decagon's per-customer support agent and Harvey's per-matter legal agent. If your business has a long-tail of entities (customers, candidates, suppliers, properties, claims) that each need continuous attention, the persistent-agent pattern is now a real architecture choice—not a research project.
The CRM is becoming an integration target, not a destination. If you are building internal AI tooling, treat Salesforce, HubSpot, and similar CRMs as data sources and execution surfaces, not platforms. Build your intelligence layer above them, with bidirectional sync. The systems-of-intelligence vs. systems-of-record split is becoming the primary architectural distinction in enterprise apps.
Per-account context is the hard part. Actively's moat isn't the model. It's the data pipeline that ingests calls, emails, CRM activity, intent signals, product usage, and external research into a coherent per-account state that an agent can reason over. If you're building anything similar in-house, plan for at least 70% of your engineering effort to land on the data plumbing, not the agent logic.
Watch the eval-and-rollout pattern. Companies winning with AI sales agents are not deploying them across the entire pipeline at once. They're starting with one named segment (e.g., dormant accounts, mid-market expansion, top-100 target accounts), measuring conversion lift against a control, and expanding. That eval discipline transfers to almost any agentic deployment.
The CIO/CRO Strategy Decoder
If you are a CIO, CRO, head of revenue operations, or head of GTM technology:
Audit how much of your sales tech budget assumes human throughput. If your sequencer, intent platform, conversation intelligence, and forecasting tools each charge per seat and assume a human running them, you have built a stack on a "people first" assumption that is being repriced in real time. The unit economics of your tech stack should match the unit economics of your revenue motion.
Run a bake-off against the per-account agent thesis. Pick 100–500 accounts that are not currently being worked. Deploy an agent system (Actively, an internal build, or a Salesforce Agentforce / Microsoft Sales Copilot variant) against them with explicit signal-driven activation. Measure conversion, meetings booked, and pipeline created over 60 days against a control set. Most enterprises that run this bake-off discover the ignored accounts produce surprising amounts of pipeline. That single data point usually justifies the architecture conversation.
Treat Salesforce as one of several execution surfaces, not the strategic decision layer. Salesforce will remain mission-critical for years. The strategic question is whether your account intelligence and decision logic should sit inside Salesforce or above it. The answer determines whether your AI investment compounds or gets locked into one vendor's roadmap.
Quota structure becomes the constraint. If you deploy AI agents that double a rep's productivity, you have to decide whether to double quota, halve headcount, or extend coverage to currently unworked accounts. Most CROs default to quota increases, which under-realizes the value. The highest-ROI move is usually expanding coverage—putting agents on the accounts you couldn't afford to staff before.
Vendor lock-in moved up the stack. The lock-in conversation used to be about CRM data export. The new lock-in is the agent's accumulated per-account context. If Actively (or any agent vendor) holds two years of contextual reasoning on your top 1,000 accounts, switching costs are real. Build contractual language for context portability into vendor agreements now, before that context becomes load-bearing.
SaaS valuations are doing your repricing for you. When Figma is down 50%, Salesforce 30%, Intuit and ServiceNow 40%, the renewal pricing leverage in your enterprise SaaS contracts is unusually strong this quarter. Use it. Multi-year renewals at flat or slightly down pricing, with explicit AI add-on rights, is a reasonable ask in 2026.
Three Concrete Moves For The Next 30 Days
-
Identify your "unworked accounts" cohort. Pull the list of named accounts in your CRM that have had no rep activity in the last 60 days but match your ICP. The size of that list is usually 5–10x your sales headcount can cover. That cohort is where the ROI of agent-based selling shows up first and most cleanly.
-
Run a 60-day controlled bake-off. Pick a 200-account test set, split it 100/100. Deploy an agent system against one half. Track meetings booked, pipeline generated, conversion. Use the same eval rubric you'd use for a model upgrade: explicit, measurable, time-bounded. Don't run a "feel" pilot.
-
Add an "agent context portability" clause to your next vendor renewal. Every AI sales tool you buy from this point forward will accumulate context that becomes harder to leave. Negotiate the right to export that context (as structured data, not just transcripts) before signing. The clause costs nothing today and protects optionality for years.
The Bottom Line
The Actively round is not interesting because of the dollar amount or the valuation. It is interesting because a $250 million startup now has reference customers like Ramp ($32 billion), Samsara, Verkada, Attentive, and Ironclad publicly endorsing the per-account agent model and reporting hard revenue impact. That is not a thesis anymore. It is a deployed pattern with measurable outcomes.
For enterprise buyers, the strategic implication is clean: the question is no longer "should we add AI to our sales process." That question was settled in 2024. The question is whether your sales architecture is built around humans pushing work through a CRM, or around agents continuously working accounts with humans intervening at decision points. Salesforce is betting it can be both. Actively is betting only one of those models scales.
The companies that get this right in 2026 will not be the ones that bought the most AI tools. They will be the ones that re-architected their go-to-market motion around the assumption that intelligence is no longer a constrained resource. That re-architecture is a CIO and CRO conversation, not a procurement one. And the window to have it without being late is narrowing.
Sources:
- Meet The $250 Million Startup Challenging Salesforce With AI Agents — Forbes, April 28, 2026
- Actively raises $45M Series B to replace human-led sales with 24/7 AI 'superintelligence' agents — Tech Startups, April 28, 2026
- Actively AI: Intelligence-Led Revenue blog post — Actively AI, April 28, 2026
- AI Giving Salesforce Boost — The Information
- Marc Benioff Says The Software Bears Are All Wrong About Salesforce — Wall Street Journal
Want to calculate your own AI ROI? Try our AI ROI Calculator — takes 60 seconds and shows projected savings, payback period, and 3-year ROI.
Continue Reading
- Google's $40B Anthropic Bet: What Enterprise AI Leaders Need to Know About the Three-Way Cloud Battle
- [OpenAI Misses Q1 Targets: Anthropic Gains 12% Enterprise Share](/article/openai-misses-internal-targets-enterprise-share-shifts-2026)
- Claude Security Finds 3x More Bugs Than Traditional Scanners
