Tech • AI • Robotics • Game
Abacus AI's New Personal Agents Are Creating a New Kind of Internet
5/10TL;DR
Abacus AI has launched cross-platform personal agents that can coordinate work across WhatsApp, Telegram, and Discord, aiming to automate team handoffs while keeping humans in charge of key decisions.
KEY POINTS
Agents handle incident response across apps
In one demo, a reliability agent for the Writer's Corner app monitored Gmail for customer issues, sent alerts to Telegram, and coordinated with a general-purpose Abacus AI agent for deeper investigation. After a user reported being unable to post, the issue was flagged P1, the highest severity, and investigated from a phone with a few text commands. The agent reproduced the bug, traced a 500 service unavailable error to a stale mock payload and a recently added feed cache flag, then produced a root-cause analysis in Markdown and PDF.
Humans approve, agents execute
After the analysis, the user authorized the fix by message, and the agent opened a pull request, removed the bad flag, and paused before deployment for explicit approval. Once approved, it merged, built, deployed, ran sanity tests, and was instructed to add the case to the regression suite. The workflow reduced the human role to three decisions: investigate, fix, and ship.
Abacus is targeting the communication bottleneck
The product is built around a simple premise: most workplace coordination already happens in chat rather than in dashboards. Instead of forcing users to leave WhatsApp, Telegram, or Slack to use AI, Abacus places an agent directly inside those channels as a normal contact. Users can create an agent in about 30 seconds by choosing a template and scanning a QR code, without handling API keys or developer tools.
Templates cover common knowledge-work tasks
Available templates include Chief of Staff, email triage, meeting prep, market brief, research monitor, sales agent, and blank custom agents. In one example, a chief-of-staff agent connected to Google Calendar, Gmail, Slack, and WhatsApp, built a prioritized to-do list, prepared meeting briefs, and checked for new tasks every four hours. It only sent messages when it detected something important.
Persistent memory is central to the pitch
Abacus is emphasizing memory across conversations and platforms, arguing that delegation fails if the AI forgets preferences. A travel-planning agent learned that a user disliked early starts, packed itineraries, tourist-heavy spots, and long restaurant lines. When later asked on another app for a trip to Tokyo, it applied those rules automatically, producing a slower itinerary built around neighborhoods rather than major attractions.
Cross-agent teamwork extends to product development
In a team demo, a product manager created a competitor watcher agent to track checkout flows at Shopify, Stripe, Square, PayPal, and Amazon. The agent was connected to a teammate's Telegram agent, allowing both users to query the same system from different messaging apps. It combined product goals from one side and engineering constraints from the other into a single product requirements document with scope, customer journey, authentication requirements, and acceptance criteria.
Broad integrations, but heavy work is escalated
The agents can connect with Gmail, Google Calendar, Slack, Jira, GitHub, Confluence, Google Drive, OneDrive, SharePoint, Teams, Outlook, Box, and Figma. Abacus says the personal agents are tuned for communication and routine work to keep costs low, while more demanding engineering or research tasks are handed off to its larger general-purpose agent.
Risks and limitations remain
The WhatsApp setup appears to rely on linked-device login similar to WhatsApp Web, meaning the agent may operate through a personal account. Abacus' open-source companion, Abacus AIBot, warns that automating a personal account may conflict with WhatsApp terms and suggests using a separate account if needed. The company also appears to have documented mainly single-agent features so far, with agent-to-agent behavior highlighted chiefly in launch demonstrations.
Competition in personal agents is accelerating
The launch comes amid a rush into always-on personal AI. OpenAI recently introduced DOTS, offering cloud-based agents with more than 4,000 app connections on a $100-a-month pro plan. Instinct raised $1 billion at a $10 billion valuation and says its agent network has already handled more than 300,000 user coordinations. Meta has also entered the field with Muse inside WhatsApp and, with Sierra and partners including Shopify, Stripe, and Walmart, is backing a personal agent protocol intended to standardize how agents interact with businesses.
Pricing may cause confusion
Abacus includes personal agents within its ChatLLM subscription, priced at $7 for the first month, then $10 a month for basic or $20 for pro. However, the personal agents page lists them under the $20 pro plan, which includes 30,000 credits a month compared with 20,000 on basic. Reviewers have cited the credit system as one of the platform's least clear features.
CONCLUSION
Abacus is betting that the next phase of AI adoption will focus less on solo assistance and more on automating coordination between people. Whether that model succeeds may depend as much on trust, platform rules, and pricing clarity as on the technology itself.
Ask a question