From Simple Automation to AI Agents at Work
Most companies already use automation, but in narrow ways: email sequences, basic chatbots, and form-based triggers that handle only one step at a time. People are still left to move data between systems, chase follow-ups, and close the gaps these tools leave behind. The emerging shift is towards AI agents at work that manage whole workflows from end to end. Instead of just answering a customer’s question, an AI agent can understand the request, retrieve the right information, update internal records, and trigger the next action. This changes automation from a series of disconnected shortcuts into a continuous, AI-powered workflow. Organisations build this capability over time by adding specific “AI skills” such as answering questions, capturing details, then updating databases or sending follow-ups. As these skills accumulate and systems become reliable enough to trust with real tasks, AI productivity tools move from being helpful extras to core parts of everyday operations.
Stacking AI Assistants: The New Developer AI Workflow
In software teams, future workplace automation is already taking shape through stacking AI assistants instead of choosing just one. A clear example is the official Codex plugin for Claude Code, which lets developers bring OpenAI’s Codex directly into Anthropic’s coding environment. Within Claude Code, slash commands like /codex:review provide a second, differently trained AI reviewer that can spot issues the first model misses. The /codex:adversarial-review command goes further by asking Codex to actively stress-test code for weaknesses, edge cases, and security risks, mirroring how serious human reviewers try to break implementations before release. Background job commands such as /codex:rescue and /codex:status allow longer-running tasks like refactoring or test generation to run asynchronously, while developers keep working inside Claude Code. The result is a dual-brain developer AI workflow where two specialised tools collaborate, improving code quality and reducing friction without forcing engineers to juggle multiple interfaces.
Beyond Engineering: AI Agents Across Email, Docs and Support
The same stacking pattern can extend beyond engineering into non-technical roles. Imagine AI agents orchestrating tasks across email, documents, and support systems. One AI triages incoming emails, classifies intent, and drafts responses. Another agent syncs decisions into your CRM, updates spreadsheets or internal knowledge bases, and schedules follow-up actions. A third might monitor support tickets, automatically answering routine questions while escalating nuanced issues to humans with a clean summary and suggested responses. Instead of one monolithic bot, you deploy several AI productivity tools, each with defined skills, chained into a cohesive workflow. This reduces the hidden manual work that currently lives between systems—copying data, reformatting documents, and chasing status updates. For employees, the job shifts from performing every micro-task to supervising and fine-tuning the orchestration: checking edge cases, handling exceptions, and improving prompts or rules so the agents can safely take on more of the repetitive workload over time.
New Skills, New Management Rules in an AI-Powered Workforce
As AI agents at work take on more responsibility, organisations must rethink both skills and management practices. Employees will need to become adept at supervising AI: writing clear prompts, defining guardrails, validating outputs, and understanding when to override automated decisions. Data literacy and basic workflow design will matter in roles that previously focused only on execution. For managers, process ownership and accountability become more complex when work is shared between humans and AI tools that collaborate. Leaders must define which steps can be safely automated, who approves AI-driven actions, and how errors are tracked and corrected. Auditability grows in importance: teams need logs of what AI agents did, why, and based on which inputs. Instead of asking whether AI will replace jobs, the more practical question is how to redesign roles so people specialise in judgment, relationship-building, and exception handling, while AI handles the repeatable, rules-based activities.
A Practical Roadmap for Malaysian Companies
For Malaysian organisations, the path to AI-powered workflows should start small and controlled. First, identify one or two high-friction processes—such as customer email triage or internal report preparation—where AI agents could reduce manual handoffs without touching sensitive approvals. Run a limited pilot with clear success metrics like turnaround time and error rates, and keep humans firmly in the loop. Next, expand by stacking AI assistants: for example, one tool drafts content while another validates data or checks compliance, mirroring how developers now combine multiple coding assistants in a single flow. Document responsibilities so staff know when to trust, review, or override AI output. Finally, scale successful pilots to team-wide rollouts with training focused on prompt design, workflow thinking, and ethical use. By moving step by step, Malaysian companies can gain the benefits of future workplace automation while maintaining safety, accountability, and trust with customers and employees.