Intent Engineering for Algerian Enterprises: Before You Deploy AI, Define What You Actually Want

Algeria is in the middle of an AI deployment sprint. Sonatrach announced its digital transformation roadmap integrating AI into upstream exploration and production analytics. Algérie Télécom is rolling out AI-powered customer service across its network. The Ministry of Post and Telecommunications (MPTIC) has positioned artificial intelligence as a pillar of Algeria’s national digitalization strategy. Banks, public agencies, and state-owned enterprises are all racing to deploy AI systems — often with more enthusiasm than clarity about what those systems are supposed to accomplish.

This race mirrors a pattern that has already played out globally, and the results should give every Algerian CTO pause. The most instructive cautionary tale does not come from a company where AI failed. It comes from a company where AI worked brilliantly — and that was the problem.

The Klarna Warning: When AI Optimizes for the Wrong Thing

In early 2024, Swedish fintech giant Klarna rolled out an AI-powered customer service agent. The numbers were extraordinary: 2.3 million conversations handled in the first month, across 23 markets, in 35 languages. Resolution times dropped from 11 minutes to 2 minutes. The CEO projected $40 million in annual savings. Klarna’s AI agent eventually replaced the work equivalent of 853 full-time employees.

Then customers started complaining. Generic answers. Robotic tone. No ability to handle anything requiring judgment. A three-year loyal customer experiencing frustration was treated identically to a first-time user with a simple question. The AI did not know that retention matters more than resolution speed. It did not know that tone mismatch is a leading indicator of churn. It did not know that some customers should be routed to humans — not because the AI is incapable, but because preserving the relationship outweighs the efficiency gain.

By mid-2025, CEO Sebastian Siemiatkowski told Bloomberg that while cost was the predominant evaluation factor, the result was lower quality. Klarna began frantically rehiring the human agents it had gutted. The 700 agents who were let go took with them the institutional knowledge that actually mattered — the understanding of which customers need patience, which situations require judgment, and which interactions build the relationship equity that drives long-term revenue.

Klarna’s organizational intent was not “resolve tickets fast.” It was “build lasting customer relationships that drive lifetime value.” But nobody made that distinction machine-readable. Not because AI failed, but because the company did not have the infrastructure to translate its actual intent into a format the AI could use.

This gap — between what AI is told to optimize and what the organization actually needs — is what the industry is now calling the intent gap. And it is about to become Algeria’s most expensive AI problem.

Algeria’s Intent Gap Is Already Forming

Consider the Algerian context. Sonatrach deploys an AI system to optimize drilling site selection. What is the objective? “Find the most productive wells” seems obvious. But Sonatrach’s actual organizational intent is more complex: balance production efficiency against environmental compliance, community relations in the Saharan regions where it operates, workforce safety metrics, and long-term reservoir management. An AI system optimizing purely for short-term production yield might recommend aggressive extraction strategies that damage long-term field viability — exactly the kind of technically correct but strategically wrong outcome that caught Klarna.

Or consider Algérie Télécom deploying AI for customer service. The measurable objective is clear: reduce call wait times, increase first-call resolution rates. But the organizational intent includes expanding broadband adoption in underserved wilayas, supporting the government’s e-governance push, and maintaining social equity in telecommunications access. An AI agent that resolves tickets fast but routes complex requests from rural customers into automated dead ends would achieve its metrics while undermining the organization’s actual purpose.

These are not hypothetical risks. They are the exact class of failure that McKinsey’s 2025 global AI survey documented: 74% of companies report no tangible value from AI investment, up from 70% the year prior — despite doubling their AI spending on average. More money flowing in, no return coming back, because organizations are deploying AI without the infrastructure to direct it toward what actually matters.

Data Spaghetti: Algeria’s Layer One Problem

Intent engineering operates at three layers. The first — where most organizations are stuck, and where most Algerian enterprises have not even started — is unified context infrastructure. This is about how data, processes, and knowledge flow to AI systems.

Algeria has a data spaghetti problem that makes the average Fortune 500 company look organized. Government data is fragmented across ministries with no unified interoperability framework. Sonatrach’s operational data lives in legacy SCADA systems, SAP environments, and departmental spreadsheets that do not talk to each other. Algerian banks operate with a patchwork of core banking systems — some modernized, some dating back decades — with customer data siloed between branches, digital channels, and regulatory reporting systems.

Deloitte’s 2026 State of AI in Enterprise report found that only 20% of global executives are fully confident their data is AI-ready. Only 14% have implemented a fully unified data strategy. In Algeria, those numbers are almost certainly lower. The country’s data infrastructure was built for reporting to supervisory authorities, not for powering autonomous AI systems that need real-time access to customer histories, operational contexts, and organizational policies simultaneously.

An AI agent trying to help an Algérie Télécom employee answer a customer question might not know that the customer has an open support ticket in one system, a pending installation order in another, and a history of service interruptions logged in a third. The data exists. It is not connected in a way the agent can access. This is the most basic layer of intent infrastructure: if agents cannot see the full picture, they cannot make good decisions.

Before any Algerian enterprise deploys AI for customer-facing operations, it needs to answer a simple question: can our AI system access, in real time, the full context it needs to make decisions that serve our organizational purpose? For most Algerian organizations, the honest answer is no.

The AI Worker Toolkit: Where Imported Tools Meet Local Workflows

The second layer is what the industry calls the coherent AI worker toolkit — how humans and agents collaborate, what tools exist, and what workflows are designed for human-agent interaction.

Algeria faces a specific version of this problem. Most AI tools available on the market were designed for organizational workflows that do not exist in Algeria. Microsoft Copilot assumes a mature Microsoft 365 deployment with well-organized SharePoint sites, clean email archives, and structured Teams channels. Salesforce Einstein assumes a CRM ecosystem with years of clean customer data. These tools were built for organizations where digital workflows have been refined over two decades.

Many Algerian enterprises are still in the early stages of basic digitalization. Paper processes remain common in government agencies. Approval chains involve physical signatures. Procurement workflows mix digital purchase orders with manual invoice processing. Deploying AI tools designed for mature digital organizations onto workflows that are still partially analog creates the same mismatch that plagued Microsoft Copilot globally: Gartner found that only 5% of organizations moved from a Copilot pilot to larger-scale deployment, and only about 3% of the total Microsoft 365 user base actually adopted Copilot as paid users.

The lesson for Algerian enterprises is not “avoid AI tools.” It is “redesign your workflows before you deploy AI into them.” Bolting an AI agent onto a workflow that was designed for paper and phone calls does not create digital transformation. It creates an expensive chatbot that nobody uses.

This requires mapping actual workflows — not the workflows that appear in official process documents, but the ones that employees actually follow. Every Algerian enterprise has informal processes, workarounds, and tribal knowledge that exist only in people’s heads. Those processes need to be documented, digitized, and structured before AI can meaningfully participate in them.

Intent Engineering: Making Organizational Purpose Machine-Readable

The third layer — and the one that almost certainly does not exist in any Algerian enterprise — is intent engineering proper. This is the discipline of encoding organizational purpose into machine-readable, agent-actionable formats.

Traditional organizational goal frameworks — OKRs, KPIs, strategic plans — were designed for humans. They encode human-readable goals that assume human judgment about prioritization, trade-offs, values, and ambiguity. When you give an employee at Sonatrach the objective “increase production efficiency,” they interpret it through layers of professional experience, safety culture, regulatory awareness, and institutional memory. They know when to push and when to escalate. They develop that judgment through years of informal mechanisms that no AI system has access to.

An AI agent needs that same judgment codified explicitly. Not as a 500-page policy document buried on a SharePoint site, but as structured parameters it can query and apply in real time. This means:

Goal translation infrastructure. Converting human-readable organizational objectives into agent-actionable parameters. “Increase customer satisfaction” is a human-readable aspiration. An agent needs to know: what signals indicate customer satisfaction in our context? What data sources contain those signals? What actions am I authorized to take? What trade-offs am I empowered to make — speed versus thoroughness, cost versus quality? Where are the hard boundaries I may not cross?

Decision frameworks codified as rules. Most Algerian organizations have well-established patterns for handling common situations. A Sonatrach field engineer knows that when two legitimate goals conflict — production targets versus safety protocols — safety wins. Always. An AI agent needs that hierarchy codified as a decision tree, not assumed as common sense.

Escalation protocols with defined triggers. Clear criteria for when agents should stop and involve humans. Not just when they are uncertain, but when the stakes are high enough that human judgment is organizationally required. In Algeria’s regulatory environment, where government oversight is a constant factor, these escalation boundaries carry particular weight.

Feedback loops measuring alignment, not just completion. Did the agent complete the task? That is table stakes. Did the agent complete the task in a way that serves the organization’s broader objectives? That second measurement is what intent engineering requires — and what Klarna lacked.

A Practical Path for Algerian CTOs

The gap between where Algerian enterprises are today and where they need to be is significant but not insurmountable. Here is a practical sequence, ordered by priority:

Months 1-3: Audit your data landscape. Before buying any AI tool, map every data source that would need to be connected for an AI system to have adequate context. Identify the gaps, the silos, and the legacy systems that cannot be easily integrated. This audit will almost certainly reveal that the prerequisites for effective AI deployment do not yet exist — and that is valuable knowledge.

Months 3-6: Document your actual workflows. Not the official process diagrams. The real ones. Interview the employees who do the work. Identify the informal processes, the workarounds, the judgment calls that happen dozens of times a day. These are the processes that AI needs to understand, and they are the processes that nobody has written down.

Months 6-9: Define your organizational intent explicitly. What does your organization actually optimize for? Not what the strategic plan says. What actually happens when goals conflict? When safety and production targets collide, what wins? When customer satisfaction and cost reduction conflict, which takes priority? These trade-off hierarchies exist implicitly in every organization. Making them explicit — and machine-readable — is the core work of intent engineering.

Months 9-12: Pilot AI with intent guardrails. Deploy AI in a limited context where you have complete data access, documented workflows, and explicit intent parameters. Measure not just whether the AI completed its tasks, but whether it completed them in alignment with your organizational intent. Iterate on the intent parameters based on what you learn.

This sequence is slow. It is unglamorous. It will not produce an impressive demo for the next ministerial visit. But it is the difference between deploying AI that works and deploying AI that works on the wrong thing — the most expensive mistake an organization can make.

The Race Algeria Cannot Afford to Lose

The race in enterprise AI is no longer about model capability. Models are converging. They are all reasonably good. What differs dramatically is whether organizations have the infrastructure to direct those capabilities toward what actually matters.

Algeria’s digital transformation push has created momentum. The national AI strategy, the investment in data centers, the emphasis on training AI professionals — these are necessary conditions. But they are not sufficient. Without intent engineering — without the organizational infrastructure that tells AI systems what your enterprise actually wants — Algeria risks repeating Klarna’s mistake at national scale. Billions invested in AI tools that optimize brilliantly for the wrong objectives.

The 700 human agents that Klarna laid off took with them institutional knowledge that was never captured, never formalized, and never made available to the AI system that replaced them. Algeria’s enterprises carry decades of institutional knowledge in the heads of their engineers, administrators, and operators. That knowledge is the most valuable AI asset they possess. The question is whether they will formalize it before they deploy the systems that are supposed to replace it.

The answer to that question will determine whether Algeria’s AI investment produces transformation or expensive disappointment.

Sources & Further Reading

• Klarna AI Agent Handles 2.3M Conversations — Klarna Press Release, 2024
• Klarna CEO on AI Quality Trade-offs — Bloomberg, 2025
• McKinsey Global AI Survey 2025: 74% Report No Tangible Value
• Deloitte 2026 State of AI in the Enterprise
• Salesforce Survey: 86% Believe Agentic AI Is Coming, 74% Not Ready
• Gartner: Only 5% Moved from Copilot Pilot to Scale Deployment, 2025
• Algeria National AI Strategy 2025-2030 — MPTIC
• Sonatrach Digital Transformation Roadmap
• Algérie Télécom Modernization Initiatives
• Google DeepMind — Five Levels of AI Agent Autonomy, 2025