CircleBack AI Note‑Taker: Workflow Automation and Recording Future

CircleBack, an AI note-taker co-founded by Ali, records and transcribes meetings, writes notes, and assigns action items. It also allows users to build automations to extract custom information from meetings and update other applications like CRMs, issue trackers, and Slack. The system can answer questions across team conversations and integrate with agents, effectively serving as a company's collective memory.

Hardware Setup and Essential Tools

Ali maintains a minimalist setup for both tools and software. A key component of his hardware is a Topre keyboard, a membrane keyboard known for its unique typing feel. Unlike mechanical keyboards with blue or brown switches, which can be loud in an open office, the Topre keyboard offers a pleasant typing experience without excessive noise. Its early key press registration means users don't need to "bottom out" each key.

Another item on his desk is a small vehicle, acquired through a Temu ad. This serves as a fidget toy during discussions or dictation. Ali also uses AirPods Max, primarily as a signal to others that he is focused and "locked in," similar to setting an "away" status on older messaging platforms.

Unexpected Uses of CircleBack

Ali uses CircleBack in several unconventional ways:

• Applicant Tracking System (ATS): He uses a custom view to track action items related to interviews. This allows him to monitor candidates, see who is interviewing whom, and review the full timeline of interactions (emails, meetings) with specific candidates. He can also check if upcoming interviews are booked or review past interview performance.
• Company and Contact Management: CircleBack helps track interactions with people and companies. For instance, he can view all contacts from Y Combinator, outstanding action items, and a full timeline of conversations and upcoming meetings.
• Meeting Preparation: CircleBack can prepare users for meetings by analyzing past emails and conversations, pulling information from calendars, and suggesting relevant topics or cool things to showcase based on shared questions.
• Quick Capture: Ali frequently uses a physical action button, set to start a CircleBack recording, to quickly capture thoughts or information without needing to interact with his phone. A forthcoming watch app will offer similar functionality, allowing users to start recordings directly from their smartwatch.

Daily Workflow and Tool Usage

Ali estimates spending 10-20% of his day with OpenClaw, as many processes run autonomously. A larger portion of his time is dedicated to coding tools, particularly for reviewing code rather than writing new code or pushing pull requests.

Telegram Agents and Customer Interaction

Ali has several Telegram agents: a main "claw" agent, and specialized agents for finance, customers, and people operations. These processes are being migrated to dedicated Slack channels to foster company-wide visibility and collaboration. For example, customer emails trigger new Slack threads for investigation, and inbound sales inquiries are handled similarly. This approach ensures everyone is aware of customer feedback, sales leads, and their outcomes.

CircleBack serves a diverse customer base, from traditional businesses to growth-stage startups and YC companies. To maintain customer engagement, the company implements a weekly "on-call" rotation where an engineer handles incoming support and interacts directly with customers. This allows for direct feedback capture, which CircleBack records. For instance, during a fireside chat with a customer, CircleBack captured feedback, and Ali can revisit the recording, jump to specific parts, or ask the AI to summarize a customer's wish list. This enables the team to track progress on customer requests and follow up with updates.

Coding Practices and AI Integration

While Ali enjoys writing code, he is consciously reducing his direct coding time to focus on strategic company growth. Thursdays are designated "no-meeting days," allowing him to focus on deep work and end-to-end shipping of projects.

He primarily uses an orchestrating agent orchestrator for coding, reviewing and refining the output. For specific tasks like copy editing or minor styling adjustments, he still manually edits code for quick iteration and contextual viewing. However, for most other development, the company has built sufficient scaffolding and infrastructure for AI agents to handle tasks end-to-end.

The bottleneck then shifts to code review, where different types of AI agents are deployed based on the pull request. These agents ensure not only code quality but also architectural consistency, product alignment, and even the consistency of copy with company standards.

Prompt Engineering and Evaluation

For quantifiable aspects of CircleBack, such as identifying action items, the team uses evaluations (evals). For example, if a task is completed during a meeting, it should not be listed as an action item to avoid redundancy.

A significant focus is on the writing style of CircleBack's notes, aiming for a human-like quality. The company has strong opinions on what constitutes good notes; for instance, they actively avoid using the word "discussed" because it adds no value. These stylistic preferences are incorporated into their evals when updating models or changing prompts, ensuring incremental improvements without introducing new issues.

Token Maximization and AI Budget

CircleBack fosters a culture of not just completing tasks but doing so in a way that improves future efficiency or output quality. This heavily involves using AI and automating operational tasks. Currently, the company operates with an unlimited budget for AI usage, trusting the judgment of its team members. While this approach might change with significant growth, the current priority is leveraging the best tools for the highest quality output. The team regularly shares new AI tools and their applications during show-and-tells.

Evolution of Workflow

Ali's workflow has significantly changed in the past six months. He spends less time in the editor, primarily using it for quick diff reviews or minor changes. Most of his coding work is now done through agents. He has also automated many operational tasks that he found less engaging as a founder, allowing him to focus on high-leverage activities and strategic decision-making.

Limitations of AI Agents

While AI agents are powerful, there are certain tasks Ali does not entrust to them:

• Sending Emails: Agents can draft emails, but never send them autonomously.
• Production Copy: AI can provide a first pass for in-product copy, but human review is essential to ensure clarity and accuracy before it goes into production.
• Data Access and Security: While agents can assist with reviewing data access and security, they are not allowed to architect these systems. The engineer or responsible party should design the architecture, with AI agents assisting in the implementation.

The Future of Recording Everything

Ali believes that as Large Language Models (LLMs) become more capable and AI agents take on more work within companies, recording everything will become a necessity. The context of conversations is crucial for AI agents to operate effectively and avoid working in isolation. The opportunity cost of not recording is already high and will only increase.

Consequently, more companies will likely default to recording all conversations and sharing relevant context across the organization. CircleBack is focused not only on making capture easy but also on ensuring customers have confidence that information is shared appropriately and securely, preventing inadvertent disclosures.

The Future of Software Engineering

Ali views the current era as an incredibly exciting time for software engineering. The boundaries of what's possible are increasingly limited by imagination rather than technical skills or bandwidth. As the cost of writing software decreases, the focus shifts from shipping speed to strategic decision-making about what to build and how to build it, emphasizing composability and interoperability of systems.

The field of software engineering is evolving rapidly, with significant changes occurring every few weeks. This dynamic environment makes it an excellent time to be a builder, especially at a startup, where there's greater freedom to experiment with new tools without bureaucratic hurdles.