THINKING FORWARD / OPEN LABS

We show how AI agents work.

Open research, reference architectures and practical experiments in connecting AI agents to real information, systems and business workflows.

Methods, sources and limitations included.

HOW AGENTS WORK

One business request.
Multiple sources. A reviewable result.

An Agent Skill defines the job and checks. MCP connects permitted tools and information sources. The agent assembles evidence, identifies exceptions and prepares the next action. A person approves consequential decisions.

How a public-service agent workflow works The agent processes a business task through six stages: business task, Agent Skill, permitted public MCP data, evidence verification, human review, business output. Public sources do not confer government transaction access. 01 / REQUESTBusiness taskA real company request 02 / METHODAgent SkillSteps, policy and stop rules 03 / DATA ACCESSMCP + public sourcesPermitted reading only 04 / CHECKSEvidence & exceptionsCitations, freshness, gaps 05 / DECISIONHuman reviewApproval before action 06 / OUTCOMEBusiness outputDecision pack and tasks Conceptual design. Government submissions require separate authorisation.
  1. Business taskA defined company request.
  2. Agent SkillInstructions, checks and stop rules.
  3. MCP + public sourcesRead only through permitted tools.
  4. Evidence + exceptionsCitations, dates and unresolved questions.
  5. Human reviewApprove consequential decisions.
  6. Business outputA decision pack and proposed tasks.
Conceptual workflow. Public-source access is read-only; submissions and changes require separate authorised integrations.

RESEARCH AREAS

From sources to executable workflows.

01

Public information & APIs

Official government portals, datasets, authentication models and access conditions.

02

MCP tools

Community and commercial MCP servers: capabilities, source attribution, maturity and limitations.

03

Entities & relationships

Licences, authorities, business activities, obligations, legal provisions, locations and deadlines.

04

Agent Skills & workflow designs

Reusable sequences that retrieve, compare, verify, prepare and escalate — with evidence and human review.

TRANSFER TO ENTERPRISE

The same engineering.
Your business systems.

The public-services research is one example of a broader challenge: turning fragmented information and rules into reliable workflows. We apply the same approach to enterprise data, operations and approvals — from architecture to implementation and team training.

Independent technical research by Thinking Forward. No affiliation with or endorsement by UAE government authorities.