From Webflow to AI: How Anita Saved $30k and Built a 10x Marketing Team
Summary of the video “She saved $30k a year canceling Webflow | Anita Kirkovska, Head of Growth & Marketing @ Vellum” by Niklas Buschner.
Anita Kirkovska, Head of Growth at YC-backed Vellum, shares how she migrated off Webflow to a custom Next.js stack powered by AI agents, built an AI-native content pipeline with skill files, and restructured her team around high-impact individual contributors rather than traditional management hierarchies. Her core insight: AI success isn't about cutting costs—it's about eliminating friction so teams can move 10x faster and produce exponentially more output.
Why Anita Left Webflow
The Friction Problem with No-Code Tools
Webflow became a bottleneck for Vellum's fast-moving startup. Publishing new landing pages took a week, engineers refused to use the visual builder, and the disconnect between Webflow and the app made PLG experiments difficult. Coding agents proved faster, so the team questioned why they were still stuck in no-code.
Webflow Didn't Adapt to the AI Era
Webflow promised AI features in 2020 but never delivered. As Vellum embraced coding agents for speed, Webflow stagnated. The platform's inability to evolve with the agentic era made it obsolete for a team betting on AI-driven workflows.
The Hidden Costs of SaaS Friction
Beyond the $30k annual Webflow fee, the real cost was slow iteration cycles, contractor dependencies, and lost engineering productivity. Eliminating friction—not just cutting subscription costs—became the priority.
The Migration Playbook
Start Small: Test Coding Agents First
Before committing, Anita's CTO spent a Sunday replicating one Webflow page using Claude and other coding agents. This low-stakes experiment proved the process could work and revealed patterns (like compressed JS) that agents struggle with.
Build a Comprehensive Migration Checklist
Webflow handles many technical SEO primitives (sitemaps, robots.txt, performance) automatically. Moving to custom code means documenting every requirement: redirects, canonical tags, image optimization, interlinking rules. Anita created a full Notion document covering all dependencies.
Partner with a Strong Technical Lead
Anita credits her CTO as essential to success. Marketers bring domain expertise and design judgment; engineers bring technical feasibility. This partnership prevented costly mistakes and kept the project grounded in reality.
Accept Design Trade-offs Today for Tomorrow's Gains
Coding agents struggle with design consistency. Rather than wait for agents to improve, Anita invested extra time upfront building design systems and rules. The bet: agents will get better at design, so the investment pays off exponentially.
The AI-Native Content Pipeline
Headless CMS + AI Agents = Content at Scale
Vellum's content lives in a headless CMS (Sanity) connected to their Next.js repo. Anita's team publishes articles by telling their AI assistant to 'publish this'—the assistant handles formatting, interlinking, cover images, and SEO rules automatically.
Skill Files: The Secret to Consistent AI Output
Skill files are markdown documents that encode process rules, API requirements, formatting standards, and interlinking logic. They're stored in the repo and accessible to AI assistants, ensuring every piece of content meets brand and SEO standards without manual review.
Humans Own Process, AI Owns Execution
For process-related decisions (what to publish, how to structure content), humans must own the thinking. For technical tasks (API documentation, integrations), AI excels. Anita's content lead spent two hours defining the publishing process with her before asking AI to document the technical implementation.
Skill Files Require Hands-On Iteration
Writing a skill file isn't a one-time task. After the first draft, the assistant uses it on real tasks. The human reviews output, updates the skill file, and repeats. This cycle builds better rules and better AI performance over time.
The Unconventional ROI of AI
ROI Isn't About Cutting Costs—It's About Eliminating Friction
Anita saved $30k by leaving Webflow, but that's not the real win. She's willing to spend 2-3x more on AI tokens and infrastructure if it means her team can move 10x faster, validate messaging in days instead of weeks, and launch campaigns at unprecedented speed.
Offboarding Low-Value SaaS, Not Cutting Spend
Vellum dropped HubSpot ($4k/month) and Webflow ($30k/year) not to save money, but because these tools slowed them down. They replaced HubSpot with Resend for email, managed entirely through AI assistants. The goal: every tool must enable speed or it gets replaced.
The Real Metric: Output Per Dollar
Instead of cost per task, measure output per dollar spent. If AI tokens cost 2x more but enable 10x output, the ROI is 5x. This shifts the conversation from 'how do we save?' to 'how do we scale?'
Enabling Non-Technical Teams to Use AI
The Illusion of Speed: Don't Skip Process Definition
New AI users often ask their assistant to 'create a skill for me' without thinking through the process first. This feels fast but produces mediocre output. Anita insists on a 2-hour workshop where the human and manager define the process together before writing any code.
Organizational Change, Not Just Tool Adoption
Using AI isn't about learning a new platform—it's about changing how work gets done. Leaders must invest in workshops, hands-on sessions, and mindset shifts. Without this, teams feel scared, confused, and disconnected from their output.
AI as Amplifier, Not Replacement
The core message to teams: AI makes you better at your job if you own the process. If you use AI prematurely without understanding your workflow, you lose agency and produce slop. Your judgment, taste, and common sense drive the process; AI executes it.
Build Confidence Through Ownership
When teams feel disconnected from AI output, they blame themselves ('I should know this better'). Leaders must instill pride in the work by ensuring teams own the process, see the results, and understand their contribution. This builds confidence to iterate and improve.
Mix Boring and Creative Work to Build Systems Thinking
Anita balances her team's weekly tasks: some repetitive (GEO article formatting), some creative (UGC campaign strategy). This mix keeps people motivated and forces them to apply insights across different contexts, naturally building systems thinking.
The Future of Leadership and Organization
Flat Organizations with AI-Enabled Individual Contributors
Traditional hierarchies exist because managers coordinate work. With AI, individuals can own entire channels (content, community, activation) and move independently. Anita now operates as a high-impact IC rather than a manager, jumping into projects where she adds the most value.
ClickUp's $1M Salary Bands Signal the Shift
ClickUp hired AI-first employees at $1M/year because they can do the work of 10 people. This signals the market's recognition that AI-native talent commands premium pay. The future rewards people who can scale their work exponentially.
High-IC Roles Require Systems Thinking
Not everyone naturally develops systems thinking. It comes from experience, education, or environment. Managers should hire for adaptability, create environments that challenge thinking, and mix boring/creative work to develop this skill in junior team members.
Direct Communication + Psychological Safety = Transformation
Anita leads by being direct about expectations while showing genuine care for each person's growth. She identifies what motivates each team member and helps them see strengths they didn't know they had. This combination builds trust and willingness to change.
AI and the Future of Work
AI Will Create More Jobs Than It Automates
While AI will automate some roles, it will make workers more productive and create entirely new types of work. History shows humans adapt well to technological change. The transition will be painful for specific roles/regions, but the long-term outcome is net job creation.
Productivity Gains Lead to New Work, Not Unemployment
People using Claude report working 10x more, not less. AI doesn't reduce work—it enables more ambitious projects. As one industry becomes highly productive via AI, other industries become relatively more expensive, creating new opportunities.
Your Agency Matters More Than the Technology
Whether you use AI well or poorly depends entirely on you, not the tool. You can be worse or better at your job with or without AI. The technology is neutral; your judgment, process, and intentionality determine the outcome.
Key Principles for AI Success
Don't Chase Trends—Build Systems
Many marketers use AI prematurely, chasing hype before understanding their own process. Anita's approach: define your system first, then use AI to execute it. This prevents dependency on AI outputs and keeps you in control.
Coding Agents Are Ready Now—Don't Wait
Anita's advice to other leaders: start using coding agents today, even if they're imperfect. Don't worry about design or edge cases—invest time to enable your team now, because agents will improve. Your competitors are already doing this.
Embrace Friction as a Learning Opportunity
The migration from Webflow to custom code was hard, but that friction forced Anita to understand SEO primitives, design systems, and process documentation. Don't avoid hard things—they build competence.
Listen to Your Peers and Stay Curious
Anita reads widely (economic perspectives on AI, founder insights), talks to peers like Elina Vera at Lovable, and attends conferences. Staying connected to the community accelerates learning and prevents isolation.
Notable quotes
If you use AI, you can be worse or better at your job. If you don't use AI, you can be worse or better at your job. It's not about AI. It's on you. — Anita Kirkovska
Don't be scared to make the trade-offs today because AI is only going to get better, right? — Anita Kirkovska
You own this. It's literally in your hands. This tech cannot change who you are. — Anita Kirkovska
Action items
- Audit your current SaaS stack for tools that create friction rather than enable speed. Identify candidates for replacement with AI-native alternatives.
- Run a small coding agent experiment: pick one landing page and have a technical partner replicate it using Claude or another agent to test feasibility and identify patterns.
- Create a comprehensive migration checklist documenting all technical SEO primitives your current platform handles (sitemaps, robots.txt, redirects, performance) before moving to custom code.
- Define your content process manually with your team before asking AI to execute it. Spend 2 hours mapping workflow, rules, and requirements in a shared document.
- Build skill files for your team's repetitive tasks: document process rules, API requirements, and formatting standards in markdown files accessible to AI assistants.
- Identify what motivates each team member and design their weekly work to mix boring/repetitive tasks with creative/strategic work to build systems thinking.
- Start hands-on AI training sessions with non-technical team members focused on process definition, not tool features. Emphasize ownership and judgment over AI outputs.
- Evaluate your organizational structure: identify high-impact individual contributor roles that could replace traditional management layers with AI-enabled autonomy.
- Read economic perspectives on AI and labor market impacts; talk to peers in other startups about their AI adoption journey; attend AI-focused conferences.
- Communicate directly with your team about AI adoption expectations while showing genuine care for their growth and identifying strengths they may not see in themselves.