The Outsourcing of Decisions

Over the past decade, Pakistan’s rise in the digital economy has been powered by one core asset: human talent. We became one of the world’s fastest-growing freelance nations, supplying code, content, design, and back-office services to global markets. Our people, not our platforms, were the advantage.

Now a new phase is arriving: Agentic AI, systems that don’t just generate outputs, but can plan, decide, and act with minimal human oversight. For Pakistan, the question is no longer whether these systems will arrive, but how we adopt them in a way that strengthens our workforce, upgrades our SMEs, and modernizes public services without creating unaccountable automation.

1) What Agentic AI Means for Pakistan’s Gig Economy

Pakistan’s freelance economy, valued at roughly $400 million, now faces competition not only from other countries, but from AI agents operating on global platforms.

Key signals are already visible:

• Oxford Internet Institute analysis shows a 20–50% drop in demand for writing, translation, basic coding, and data entry after widespread adoption of tools like ChatGPT.
• Upwork reported a 60% year-on-year increase in AI-related projects; freelancers who use AI tools in their workflow tend to earn more.
• AI-powered freelancers can submit 50+ proposals per day, increasing volume-driven competition.
• Changing client expectations such as faster delivery, lower pricing, and broader skill coverage are increasingly assumed.

This is where Pakistan’s adoption strategy matters. If our freelancers only “use” AI casually, they’ll be outpaced by those deploying agents systematically. But if we mainstream agentic workflows like proposal automation, portfolio generation, rapid prototyping, and compliance-friendly toolkits, we can shift from low-margin tasks to higher-value AI-enabled services.

2) What It Means for Tech SMEs

The pressure extends beyond freelancers. SMEs that deliver software services will increasingly compete in a market where clients expect agentic acceleration.

• Gartner predicts 33% of enterprise software applications will include agentic AI by 2028, up from less than 1% in 2024.
• In the US, 79% of organizations report some level of AI agent adoption; 96% globally plan to expand AI agent use in the next 12 months.
• Since 2023, nearly 600 new AI-related skills have appeared in job postings—about one-third of all new digital skills added in the past decade.

For Pakistani SMEs, this creates a clear adoption roadmap: integrate agents into delivery pipelines (requirements-to-code, test automation, documentation, DevOps, support), retrain teams for hybrid roles (AI product owners, AI QA, AI security), and build credibility through governance-by-design (audit logs, model cards, data handling policies).

SMEs that do this early can become partners for global firms. Those that don’t will face margin compression and “race-to-the-bottom” pricing.

3) From Chatbots to Decision-Makers: The Public Sector Adoption Moment

For government, the biggest shift is not AI-generated text, it’s AI-executed decisions.

• 2023–2024 (Generative AI): A department uses AI to draft letters, translate documents, or summarize cases. A human approves. Governance is manageable.
• 2025 (Agentic AI): A ministry deploys an agent that can process thousands of applications, flag anomalies, prioritize cases, trigger notices, recommend approvals/rejections, or initiate follow-ups, often across multiple systems.

This is precisely where Pakistan can benefit: agentic systems can reduce backlogs, improve service delivery, and standardize decision processes. But it is also where adoption becomes risky if guardrails are missing.

If an agent can decide at scale, then public sector adoption needs clear rules on:

• where human sign-off is mandatory,
• what explanations citizens can demand,
• how decisions are logged and audited,
• who is accountable when automation causes harm.

Pakistan’s draft Personal Data Protection Bill (2023/25) includes Section 29, granting individuals the right not to be subject to decisions made solely through automated processing. However, enforcement and operational clarity remain uncertain. Meanwhile, the National AI Policy 2025 sets ambition on skills and innovation, but is still light on agency-specific governance such as liability, auditability, and deployment standards for autonomous decision systems.

This matters because without a trusted governance layer, public institutions may either:

• avoid adoption (fear of controversy and legal ambiguity), or
• adopt informally (deploy tools without accountability structures), which is worse.

A practical example: if an AI system used in an insurance or relief process rejects flood-related claims based on automated profiling with no human review, what is the citizen’s path to challenge the decision? Without defined safeguards, adoption erodes trust, even if efficiency improves.

How Pakistan Can Adopt Agentic AI Responsibly

Pakistan doesn’t need to slow down adoption. It needs to adopt with design discipline.

Technical and institutional priorities

• National AI testing labs to audit systems before use in sensitive services.
• An AI registry documenting deployed models/agents, purpose, owner, scope, and limitations.
• Capacity building at scale across public and private sectors in AI deployment, evaluation, auditing, and risk management.

Policy priorities

• Mandatory human review for high-stakes domains (benefits, policing, finance, health, hiring).
• Right to explanation in local language for AI-influenced decisions.
• AI system identification so deployed agents are traceable to responsible entities.

Canada’s Algorithmic Impact Assessment and Singapore’s AI Verify show that innovation and governance can move together, not sequentially.

Conclusion

Agentic AI will reshape how work is won, how software is built, and how public services are delivered. For Pakistan, the opportunity is not just to “use” agentic AI, but to build competitive capability around it, so freelancers upgrade, SMEs scale, and government modernizes responsibly.