Circleback CEO Ali Haghani: Recording Company Meetings Will Become The Norm
Ali Haghani, co-founder of CircleBack, shares his unique hardware setup and delves into how his AI notetaker streamlines business operations, from interview tracking to customer support. He discusses the shift from manual coding to AI-orchestrated development, the nuances of prompt engineering, and his vision for why recording meetings with AI is becoming essential for leveraging LLMs and agents effectively in enterprises, ultimately reshaping the future of software engineering.
Ali Haghani, co-founder of CircleBack, a Y Combinator Winter 2024 alumnus, provides an in-depth look at his work environment and the transformative role of AI in his company and personal workflow. CircleBack itself is an advanced AI note-taker designed to capture, organize, and automate insights from conversations. It records and transcribes meetings, generates notes, assigns action items, and can build automations to integrate custom information with other applications like CRMs, issue trackers, and Slack. Essentially, it functions as a “company brain” that retains all conversational context, allowing users to query information and connect AI agents to it.
Hardware and Workspace Philosophy
Haghani advocates for a minimalist setup, both in terms of physical tools and software. A notable piece of his hardware is a Topre (Tokyo Precision) membrane keyboard. Despite the challenge of acquiring it and its Japanese-only instruction manual, he praises its quiet operation and the tactile experience of its “early registration of each key press,” distinguishing it from noisier mechanical keyboards like those with blue or brown switches. Other items include a fidget toy, acquired unexpectedly through a “hypnotizing” Temu onboarding process, which serves as a tactile aid during discussions, and AirPods Max, primarily used to signal a “locked-in” status during focused work.
Unconventional Applications of CircleBack
Haghani demonstrates CircleBack’s versatility beyond standard note-taking:
- Applicant Tracking System (ATS): He uses a custom view to manage action items related to interviews, track candidate progress, and monitor interactions, effectively using it as a personalized ATS. It provides a full timeline of candidate touchpoints, ensuring follow-ups and performance reviews are logged and accessible.
- Company Relationship Management: CircleBack serves as a CRM for tracking interactions with external entities like Y Combinator. It consolidates conversations, outstanding action items, and upcoming meetings, providing a unified historical and future view of engagements.
- Personalized Meeting Preparation: For this interview, Haghani leveraged CircleBack to review past email exchanges, previous conversations, and calendar entries to prepare for potential questions and identify relevant showcases.
- Rapid Information Capture: He utilizes the Apple Watch’s action button (and an upcoming watch app complication) to initiate recordings quickly, allowing for hands-free capture of thoughts or discussions while moving around, eliminating the need to interact with a phone.
AI Agents and Workflow Transformation
Haghani estimates spending 10-20% of his day interacting with “OpenClaw” (referring to his AI agents, likely a custom setup with a Telegram interface) and the rest on traditional coding tools, though increasingly on code review rather than direct writing. His goal is for agents to be “always on and just working.”
He has developed several personal Telegram-based agents for specific functions: “Main Claw,” “Finance,” “Customers,” and “People Ops.” These are being migrated to dedicated Slack channels within the company to enable collective effort and transparency regarding customer inquiries and sales leads. This approach ensures everyone has visibility into customer needs and sales opportunities. CircleBack also facilitates weekly “on-call” rotations for engineers to handle customer support, providing direct customer interaction. An example highlights how CircleBack captured customer feedback during a fireside chat, allowed querying for specific “wish list” items, and could even draft follow-up emails confirming shipped features.
Shifting Role of a Founder and Engineer
While Haghani enjoys writing code, he is consciously reducing his direct coding time to address company bottlenecks related to hiring and product decisions, as the engineering team is already strong. Thursdays are his dedicated “no meeting” days, reserved for deep work and end-to-end feature development.
His current coding flow primarily involves using an “orchestrating agent orchestrator,” with manual editing reserved for quick diff reviews, copy adjustments, or minor styling changes. The extensive scaffolding and infrastructure built at CircleBack enable AI agents to handle most development tasks end-to-end. The bottleneck has shifted to code review, which is also augmented by AI agents that assess code quality, architectural consistency, product alignment, and even copy.
Principles of Prompt Engineering and Evals
CircleBack employs robust evaluation systems (“evals”) for quantifiable outcomes and specific criteria. For instance, an action item should only be generated if it’s truly pending; if a task like “I’m going to send you this right now” is completed during the meeting, it should not be logged as an action item.
A significant focus is placed on the writing style of CircleBack’s output, aiming for human-like prose. The team holds strong opinions on “what good looks like,” such as explicitly avoiding generic and valueless phrases like “discussed” in notes, as “obviously, things were discussed, it’s a meeting.” These specific stylistic rules are integrated into their evals, ensuring incremental improvements without introducing new issues (e.g., making overviews better but making topics too long).
“Tokenmaxxing” and AI-First Culture
CircleBack fosters a culture of not just completing tasks, but doing so in a way that improves future efficiency. This involves extensive use of AI and automating mundane “ops tasks” that no one should be doing manually. The company currently operates with an “unlimited budget for using AI,” trusting the team’s judgment to leverage the best tools for high-quality output. Internal “show-and-tells” are common, where team members share newly adopted tools and their usage.
AI Agent Limitations
Despite their extensive use of AI, certain boundaries are strictly maintained:
- Agents are never permitted to send emails autonomously, though drafting emails is acceptable.
- Product copy can undergo a first pass by an agent, but requires human review before production.
- Anything involving data access and security is handled cautiously. Agents can review, but are not allowed to architect solutions independently; human engineers must architect, and agents can then build.
The Imperative of Recording Meetings
Haghani posits that as Large Language Models (LLMs) and AI agents become more sophisticated, recording meetings will become a necessary practice for companies. The core thesis is that context is paramount; without recorded conversations, AI agents operate in isolation, lacking crucial information about past discussions or customer feedback. The “opportunity cost of not recording is already pretty high and it’s only going to become higher and higher.” Companies will increasingly default to recording and sharing conversational context across the organization. CircleBack prioritizes not only easy capture but also providing customers with confidence that captured information is shared appropriately, preventing inadvertent disclosure and encouraging comprehensive usage.
The Future of Software Engineering
Haghani expresses immense excitement about the current state of software development, where “the bounds of what’s possible is now more so your imagination, not so much the skills that your team or you have, or the bandwidth that you have.” As the cost of writing software decreases, the emphasis shifts from speed of shipping to the strategic decision of what to build and how to build it to ensure systems can be composed and interoperate effectively. He finds the rapid evolution of the field “seemingly like every few weeks” invigorating, particularly for builders in startups who can leverage the latest tools without bureaucratic constraints.