Разработка ПО
Менеджмент
Для компаний
Для стартапов
Архитектура ПО
The company cut 70% of Salesforce licenses, but bills grew by 80%. An AI-VP of marketing now launches campaigns without human involvement. A team of 2 people and 20+ agents replaced a staff of 15 employees. Here is what this means for your business.
Оглавление
What happened
The SaaStr team published the third episode of The Agents podcast, where two people and more than 20 AI agents run a company with tens of millions of dollars in revenue. The key event of the week: their AI-VP of marketing named 10K crossed the line for the first time — it began independently generating, segmenting, and preparing campaigns for launch without human involvement. Every Saturday morning, the system analyzes data from Salesforce, ticketing, and sponsor analytics, selects the three best ideas, and prepares an email campaign to “ready to send” status.
In parallel, a dramatic incident unfolded with Marketo: the platform stopped processing unsubscribes, and users kept receiving emails after requesting deletion three times. On the tenth day, without getting a response from support, the team built its own unsubscribe handler on Replit in 20 minutes, bypassing the broken Marketo API directly. This took less time than it took the vendor to acknowledge the bug existed.
How this is useful for business
The consumption model replaces the seat model. Salesforce reduced the number of licenses by 60-70%, but the bill grew by 80% — from $12-16K to $22K per year. The reason: payment shifted to API calls, agent actions, and Agentforce consumption. The company does not buy seats in the system — it pays for the real work the system performs. This is a fundamental shift in B2B software pricing.
Data stops being hostage to the interface. The team does not have a login in Salesforce, but works with data through the API every day. Information that previously required navigating an inconvenient UI is now available through natural language via an AI agent. The CTO of Owner used this approach to analyze transcripts of hundreds of calls and identify the real reasons for won and lost deals — at the level of product features, not subjective manager assessments.
How to make money from this
The architecture of headless integrations is becoming the main product. When data from CRM, analytics, ticketing, and finance is combined into a single surface for AI agents, demand emerges for specialists capable of building it. Aaron Levie named a new role: a $500K-$1M per year operator — a person who deploys agents, connects APIs, and fixes what is broken. This is not an engineer, but a technically literate enough specialist who loves to build.
The support market for AI companies is empty and enormous. Replit, Clerk, Recall, and dozens of other hot AI startups do not have normal support. The SaaStr team states: AI companies have forgotten how to do support, although support is the main use case for AI agents according to Salesforce and Agentforce data. The best vendors are now winning thanks to Forward Deployed Engineers even for the SMB segment.
Business ideas
1. A stealth churn monitoring service for B2B platforms. Create a system that tracks DAU/WAU/MAU of paying clients and signals when a user has not logged in for 30, 60, 90 days. Sell subscriptions to vendors who want to know real churn, not just cancellations. Model: $500-2000/month for connecting to a client’s API.
2. An agency for building N=1 applications. Every company has 20+ tasks that eat up employees’ time: distributing parking passes, processing requests, internal directories. Build such applications in one or two days on Replit, Lovable, or analogues. Position it as a replacement for hiring a part-time FTE. Cost: $2000-5000 per application, $1000/month for support.
3. Consulting on migration to consumption-based models. Help B2B vendors move from licenses to usage-based payment. Consulting includes reworking pricing, integrating consumption analytics, and configuring triggers for upsell. Project cost: $15-50K.
4. AI-VP marketplace for SMB. Package ready-made agents for typical functions: VP Sales, VP Marketing, VP Operations. A small business cannot hire a vice president for $200K, but can pay $500-2000/month for an AI version. Create integration templates with popular CRM and marketing platforms.
5. A “support as a service” service for AI companies. Offer vendors a team of live specialists who close tickets in 15-60 minutes. AI companies do not want to build support, but understand that without it they lose customers. Model: retainer $5000-20K/month for SLA.
Risks and limitations
Dependence on API access. If a vendor changes the API or restricts access, your headless stack will fall apart. Marketo has not updated its API since 2014 — this is not an isolated case, but a systemic problem of legacy platforms.
Regulatory risks with autonomous sending. The AI-VP of marketing in the case stopped before auto-sending not because of technical limitations, but because of legal ones. CAN-SPAM and GDPR require human control over email campaigns. Autonomous sending without a human in the loop is an area of legal risk.
Stealth churn works both ways. If you build a business on agent-based sales, your clients can stealth-abandon you in exactly the same way. DAU/WAU metrics are becoming critical for B2B companies already now.
7-day action plan
Day 1-2: Audit your tech stack. List all the platforms you pay for and ask each employee which of them they have logged into in the last 30 days. Compare the list with real usage — find Notion, Canva, and other subscriptions that people have grown tired of without formally canceling.
Day 3: Connect consumption analytics to your CRM. If you are a vendor — start tracking client DAU/WAU. If you are a client — calculate the cost of one API call and compare it with the license price. Understand which side the real benefit is on.
Day 4: Find one manual task in your company that takes more than 2 hours a week. This is a candidate for an N=1 application. Describe the process in natural language and try to build a prototype in Replit or Lovable in one day.
Day 5: Study the market for Forward Deployed Engineers in your niche. This is a scarce resource, and demand is growing faster than supply. If you are looking for such a specialist — start monitoring LinkedIn and specialized communities now.
Day 6: Check the unsubscribe mechanism in your product. If you are a vendor — make sure an unsubscribed user really stops receiving communications. If you are a client — check whether unsubscribing works on all platforms you use.
Day 7: Make a decision on one of the “dead” tools found last week. Cancel the subscription or move the team to regular use. Record the result and the savings.
Original news: SaaStr · See other news in the news section.
