How AI Will Change 'Business Operations' in 18 Months: The Enterprise Adoption Roadmap Shown by Anthropic × HubSpot
"How will AI change business in 18 months?"—At the main session of INBOUND 2025, HubSpot's Yamini Rangan and Anthropic's Dario Amodei discussed exponentially growing model capabilities, the reality of enterprise adoption, and the conditions for widespread use centered on "safety and trust." Below, we organize the latest facts and key points from their remarks, translating them into an adoption roadmap that companies can implement over the next 18 months.
1. The 'Tension' of Exponential Growth: Where Common Sense Gets Betrayed
Amodei emphasized that we should think between two voices: "the common-sense view that it should stop" and "what if it keeps growing anyway." His assessment is that if current growth continues for another 1 to 3 years, AI could reach discoveries (in medicine, life sciences, etc.) that surpass the "frontiers of human knowledge." The session was framed not as "excessive certainty" but as a probability bet, illustrating a structure where the emergent capabilities of models become the source of enterprise value. The full picture of his remarks can be confirmed through official session information and published videos.
1-1. From "Emergence" to "Experimental Verification"
Over the next 18 months, a three-stage rollout of theoretical possibility → business process experimentation → limited production will become mainstream. Coding, pharmaceutical R&D, and financial credit/fraud detection are the leading sectors for this ripple effect. At INBOUND, it was repeatedly mentioned that "exponential diffusion in the developer domain will act as a catalyst for other departments."
2. Behind the "Super Hyper-Growth": Facts about Anthropic
In the week leading up to INBOUND, Anthropic completed a $13 billion Series F round, reaching a post-money valuation of $183 billion. It is reported that their run-rate revenue accelerated from approximately $1 billion at the start of the year to over $5 billion as of August. This is driven by a shift in focus toward enterprise use (API/B2B) and the rapid growth of Claude Code.
2-1. Claude Code is a "Breakthrough for Adoption"
Claude Code saw rapid expansion in developer adoption and revenue run-rate after its release. Through monthly usage tracking and the publication of case studies, it followed the classic path of internal PMF (Product-Market Fit) leading to external sales. Market research also confirms the path of crossing the chasm from "developer-focused to spreading to other departments."
3. Shifting the Center of Gravity from "Consumer to Enterprise": The Adoption Blueprint Shown by HubSpot
HubSpot announced over 200 new features at INBOUND 2025, along with an AI-era growth playbook called "The Loop" (Express → Tailor → Amplify → Evolve). With Data Hub and Breeze Agents (a suite of task-specific agents), they are positioning themselves as a front-office OS that assumes a "human × AI" hybrid operation. For SMBs, this serves as an implementation foothold for "where to start."
3-1. Why does it work even for SMBs?
The keys are contextualizing internal data (Smart CRM/Data Hub) and AI components that can be reused across departments (Breeze Agents/Studio/Marketplace). A pricing structure that facilitates a small start and easy expansion—combining seat-based billing with usage-based billing—has also been presented.
4. How to Ensure "Safety and Trust": Threats that Actually Occurred and Countermeasures
In just the last few weeks, Anthropic announced and blocked large-scale unauthorized use by attackers suspected of state involvement. 17 organizations, including medical and government entities, were targeted, and it was reported that Claude Code was misused in some stages of external threats. The company has implemented enhanced detectors, updated terms of service, and blocked the accounts in question.
4-1. Browser Agents are a "Phased Adoption"
Anthropic has limited Claude for Chrome to a pilot of 1,000 users and is currently verifying defenses against prompt injection. They have clearly stated test conditions such as "do not input confidential data" and have published permission-based authorization designs and warnings. The decision to avoid a hasty, full-scale rollout can be considered a best practice for safety in enterprise adoption.
4-2. Practical Points for Governance
Permissions and Boundaries: Minimize the scope of access for browser extensions and RAG.
Injection Resistance: Dual-layer countermeasures for both models and middleware that do not unconditionally trust instructions derived from content.
Abuse Monitoring: Anomaly detection of account behavior and prompt blocking. Continuous sharing and learning of external cases.
5. 18-Month Roadmap: What to Do Now and in What Order
5-1. 0–90 Days: Establishing the Foundation
Use Case Selection: Narrow down to three cases where ROI is easy to verify, such as code assistance, summarization/drafting, and search/QA (it is safe to make one of these for developers).
Data Preparation: Connect to HubSpot Data Hub/Smart CRM or internal data lakes and clarify the classification of personal information and confidential data.
Safety Gates: Establish model selection (corporate policy), privilege minimization, audit logs, and injection countermeasure policies.
5-2. 3–12 Months: The Run-up to Scaling
Quantification of Developer Productivity: Measure commit counts and PR lead times with tools like Claude Code, and expand to each team.
Departmental Expansion: Gradually deploy Breeze Agents to Sales (proposal drafting), Marketing (personalization), and CS (self-resolution).
Limited Implementation of Browser Agents: Test automatic clicking systems limited to internal sites and test data, and create a dashboard for safety metrics (e.g., injection success rate).
5-3. 12–18 Months: The Core of Business Transformation
Productionalizing 'Virtual Colleagues': Semi-automate tasks involving multiple SaaS integrations, such as applications, credit checks, and incident response, using workflows + agents.
Model Diversity and Cost Optimization: Compare inference unit prices, SLAs, and regions to adopt a multi-vendor approach.
In-house Risk Standards: Operate model behavior auditing, derailment detection, and PII leakage prevention as internal standards.
The realistic order is 'Developers → Sales/Marketing → CS → Back Office → Regulatory Departments.' Success experiences with developers serve as convincing evidence for expansion to other departments.
6. Investor Perspective: Drivers of Valuation Expansion and 'Cool-Headed Assumptions'
Drivers: Enterprise API demand, expansion from coding to other departments, and differentiation through safety and governance compliance.
Premise: The reality that infrastructure spending is heavy and security incidents can happen at any time. Anthropic's approach of starting with limited pilots and then scaling in stages is rational for simultaneously optimizing product value and reputation.
7. Summary
The key to applying the 'tension' of exponential functions to actual management is to start with early wins among developers, make data preparation and safety governance a prerequisite, and carve out an 18-month 'limited to expanded' roadmap through operational designs for continuous learning like The Loop.