This post explains how a team of software agents can take a web project from the first idea to a live website on the Internet. It joins two units: the agent ideas of Unit 13 (Emerging Trends in ICT) do the work of Unit 10 (Web Development).
The post has five parts:
- The two lessons it joins: software agents and multi-agent systems, and the Unit 10 web steps (sections 1 and 2).
- How the team is built: why one agent is not sufficient, the agent team and how the agents communicate (sections 3 to 5).
- How the website goes live: deployment from localhost to a live server, and maintenance after it (section 6).
- How the team is run: centralized or decentralized control, Waterfall or Agile, and the advantages and risks (sections 7 to 9).
- How the exam tests it: past paper questions on agents (section 10).
1. Software agents and multi-agent systems
A software agent works for a user, and a multi-agent system is a team of agents that solves a problem together.
Software agent
A software agent is software that works for a human user or for another program. The Teachers' Guide says that it works "autonomously and continuously in a particular environment" (Teachers' Guide 13.2, page 110) (ගුරු අත්පොත 13.2, පිටුව 111). The Sinhala Guide writes this as "ස්වායත්ත සහ අඛණ්ඩ".
The two Sinhala books use different words for "agent". The Sinhala Teachers' Guide says "මෘදුකාංග කාරක" and "බහු කාරක පද්ධති" in its contents and definitions. But its level title on the same page says "නියෝජිත" (ගුරු අත්පොත 13.2, පිටුව 111). The Sinhala Resource Book says "මෘදුකාංග නියෝජිත" and "බහු නියෝජිත පද්ධති" (Resource Book 13.2, පිටුව 162). Learn the English word "agent", because it is the same in both media.
The Resource Book gives these characteristics of a software agent (Resource Book 13.2, page 145) (Resource Book 13.2, පිටුව 162):
- Autonomous: it makes decisions without the human user. The Sinhala book calls it "ස්වාධීන බව (independence)".
- Proactive: it starts actions on its environment. The Sinhala book calls it "ක්රියාශීලීත්වය".
- Reactive: it reacts to events in its environment. The Sinhala book calls it "ප්රතික්රියාශීලීත්වය".
- Cooperative: it works with other agents and with humans. The Sinhala book calls it "සහයෝගයෙන් වැඩ කළ හැකි බව".
- Able to learn: it learns from observation, experience and data. The Sinhala book calls it "ඉගෙන ගත හැකි වීම".
- Social ability: it communicates with others and obeys the rules. The Sinhala book calls it "සමාජ හැකියාවන් සහිත වීම".
The Sinhala Resource Book has a misprint in this list. It prints the English word as "corporativeness". The correct word is "cooperativeness" (Resource Book 13.2, පිටුව 162).
These characteristics make an agent different from a usual object in a program (Resource Book 13.2, page 145) (Resource Book 13.2, පිටුව 162).
Multi-agent system (MAS)
A multi-agent system is a group of agents that works as one unit. The agents work together because one agent cannot solve a complex problem alone (Resource Book 13.2, page 145) (Resource Book 13.2, පිටුව 163). In a MAS, each agent (Resource Book 13.2, page 146) (Resource Book 13.2, පිටුව 163):
- is autonomous: the Sinhala book says "ස්වාධීන වීම (independence)",
- has only a local view: it knows only its own part of the work ("දේශීය දර්ශනයක් පමණක් සහිත වීම"),
- acts in a decentralized manner ("විමධ්යගත ආකාරයකට කටයුතු කිරීම").
Figure 5.4 in the Resource Book shows a user, an interface agent, a broker agent and some information agents (Resource Book 13.2, page 146) (Resource Book 13.2, පිටුව 164). The Sinhala figure labels them "පරිශීලක", "නියෝජිත අතුරුමුහුණත", "තැරැව්කාර නියෝජිත" and "තොරතුරු නියෝජිත". Remember this figure. The web project team in this post uses the same shape.
2. Web development steps in Unit 10
A web project in the syllabus has six steps, and only the last step puts the website on the Internet.
- Find the need and the type of website (10.1).
- Analyse the user requirements (10.2).
- Make the pages with HTML (10.3, 10.4).
- Change the appearance with CSS (10.5).
- Make dynamic pages with PHP and MySQL (10.7).
- Publish the website and maintain it (10.8).
Many students stop at step 5. The website works on localhost. But localhost is only the developer's own computer. Nobody on the Internet can see it. Step 6 is the step that makes the website live.
3. Why one agent is not sufficient
One agent is not sufficient because a complete website needs too many different skills for one job.
The skills are requirements, design, HTML and CSS, PHP, SQL, testing, security, servers and domain names. One agent with all these tasks has a very large job. It can easily forget a part or mix up the parts.
When the work is divided, each agent has a small and clear job. This is the "local view" from section 1. Each agent is good at one thing. The team solves the full problem.
4. The web development agent team
The team has ten agents, and each agent has one job. The numbers show the syllabus competency level that the job comes from.
- Interface agent. It talks to the client (the human user). It collects the client's needs and shows the results to the client. This is the same role as the interface agent in Figure 5.4.
- Orchestrator agent. It makes the plan. It gives each task to the correct agent. It collects the outputs and checks them. This is the same role as the broker agent in Figure 5.4.
- Requirements agent (10.2, 7.5, 7.6). It writes the functional and non-functional requirements. For example: "A student can register", "A page opens in less than 3 seconds".
- Design agent (7.7, 8.4, 8.5, 8.6, 8.7). It designs the page structure, the navigation and the database. It makes the ER diagram and normalizes the tables.
- Front-end agent (10.3, 10.4, 10.5). It writes the HTML pages and the CSS style sheets.
- Back-end agent (10.7). It writes the PHP code, the forms (
GETandPOST) and the SQL for the MySQL database. - Test agent (7.8). It tests each unit, then the full system. It sends each error back to the agent that made it.
- Security review agent (6.11). It looks for security problems. For example: passwords stored as plain text, or form input that goes directly into an SQL statement.
- Deployment agent (10.8, 7.9). It puts the website on a live web server.
- Monitor and maintain agent (10.8). It watches the live website, makes backups and reports problems.
5. How the agents communicate
The agents communicate in two ways: messages and a shared workspace. They do not share one brain.
- Messages. The orchestrator sends a task to an agent. The agent sends back a result or a problem. This is the "cooperative" and "social ability" characteristic.
- A shared workspace. All the outputs go into one place: the requirements document, the designs, the source code and the test results. Usually this is a version control store. Each agent reads the parts that it needs. Each agent writes only its own part.
The shared workspace lets each agent keep a local view, while the full team still sees one project. It also keeps a history. If an agent makes a bad change, the team can go back to the last good version.
6. Deploy a website from localhost to live
A website goes live when the deployment agent publishes it on a web server, and it stays useful only when the monitor agent maintains it. Many notes leave out this part. The Resource Book gives two ways to publish a website (Resource Book 10.8, page 123) (Resource Book 10.8, පිටුව 135):
- Local publishing. The developer's own computer is the web server. It needs a good Internet connection and a static IP address from the ISP. The router must send traffic to port 80. The firewall must let the web server use port 80. It also needs a web server program, such as Apache. The Sinhala book says "ස්ථිතික IP ලිපිනයක්" for a static IP address, "කෙවෙනි 80" for port 80 and "ගිනි පවුර" for the firewall.
- A web hosting company. The company keeps the web server, the connection and the security for the website owner. The Sinhala book calls it a "සත්කාරක (hosting)" company.
The Teachers' Guide also lists free and paid hosting, and shared and dedicated hosting (Teachers' Guide 10.8, page 96) (ගුරු අත්පොත 10.8, පිටුව 96). The Sinhala Guide says "නොමිලයේ සහ මුදල් ගෙවා" for free and paid, and "සාමූහික සහ එක් පාර්ශ්වයකට පමණක් සීමා වූ" for shared and dedicated.
The deployment agent does these steps, one after the other:
- Select the hosting. Shared, dedicated or another type. This decision costs money, so a human approves it.
- Get a domain name and connect it to the server with DNS (6.9).
- Prepare the server. Install the web server, PHP and MySQL.
- Keep the secrets safe. The database password goes into a settings file on the server. It does not go into the source code.
- Make the live database. Make the tables from the design. Do not copy the test data.
- Copy the website files to the server.
- Turn on HTTPS with a certificate, so that the data between the browser and the server is encrypted (6.11).
- Do a smoke test. Open the live address. Register a test user. Make sure that the main pages work.
- Keep a way back. If the new version fails, go back to the last good version.
The deployment agent can also use the deployment methods from Unit 7 (7.9). For example, it can put the new version on a small part of the users first (pilot), or it can change all at once (direct).
After the website is live
Publishing is not the end, because a website must be maintained regularly (Resource Book 10.8, page 123) (Resource Book 10.8, පිටුව 135). The Sinhala book says "වෙබ් අඩවිය නිරතුරු ව පවත්වාගෙන යා යුතු ය". The monitor and maintain agent does this work:
- It makes scheduled backups.
- It applies software updates that fix bugs and security problems.
- It finds broken links and pages (Resource Book 10.8, page 123) (Resource Book 10.8, පිටුව 135).
- It collects site statistics, for example the number of visitors (Resource Book 10.8, page 123) (Resource Book 10.8, පිටුව 136). The Sinhala book calls them "සංඛ්යාලේඛන (statistics)".
- It watches the performance: download speed, concurrent users and errors (Teachers' Guide 10.8, page 96) (ගුරු අත්පොත 10.8, පිටුව 96).
When it finds a problem or a new need, it sends it to the orchestrator. Then the cycle starts again.
7. Centralized or decentralized MAS
A web project team works best as a centralized MAS, although the Resource Book describes a MAS as decentralized.
Use them like this. In a decentralized MAS, no agent controls the others. The agents agree among themselves. In a centralized MAS, one agent (a broker or an orchestrator) controls the flow. Each agent is still autonomous in its own task.
A web project has a clear order of stages: requirements first, then design, then code, then test, then deploy. So a centralized design with an orchestrator is easier to control. This is why the team in section 4 has an orchestrator. The 2023 paper also asks about a MAS with a broker agent and centralized control (2023 (held 2024) Paper II, B, Q7(b)(iv) and (v)).
8. Waterfall or Agile with agents
The agent team can follow different SDLC models (7.3), and only the orchestrator's plan changes between them.
- Waterfall. Each agent finishes its stage. Then the next agent starts. The orchestrator moves the work from one stage to the next.
- Agile. The team makes a small working part, puts it live, gets feedback from the client, and does the next small part. Agents are fast, so many short cycles are possible.
In both models, the agents are the same.
9. Advantages and risks of multi-agent systems
A MAS gives specialists and parallel work, but it also brings risks that need controls.
Advantages
- Each agent is a specialist in one job.
- Agents can work at the same time. For example, the front-end agent and the back-end agent work in parallel.
- When one agent fails, the team replaces or fixes only that agent. The other agents continue.
- A new agent is easy to add. For example, an agent that translates the website into Sinhala and Tamil.
Risks
- An agent can give wrong information with confidence. A second agent (the test agent) must check the work.
- When agents do not cooperate, the work stops or the parts do not join correctly.
- An agent that can deploy can also break the live website or delete live data.
- A secret, such as a database password, can leak if an agent puts it in the wrong place.
10. Past paper questions on agents
Agent technology (13.2) appeared in every new-syllabus paper from 2019 to 2026, mostly in Paper II, Part B, Question 7. The questions often ask students to draw a MAS diagram, to find the interface agent, or to compare autonomy.
The 2023 mark of 10 is an estimate, because no official marking scheme gives it. The 2022 and 2024 marks come from answer keys that the Department of Examinations did not publish. So the real marks for these three years can be a little different.
The 2026 paper is very close to this post. A software company uses a MAS to automate the SDLC for a client's project (2026 Paper II, B, Q7(b)). It has four agents: requirements analysis, system design, implementation and testing, and deployment and maintenance. This post does not give the answer. Draw the diagram first. Then check the answer against sections 1, 4 and 7.
More practice:
- 2025 Paper II, B, Q7(b): MAS diagram, advantages over one big program, risks of wrong information.
- 2023 (held 2024) Paper II, B, Q7(b): cooperation, broker agent, centralized control.
- 2021 (held 2022) Paper II, B, Q7(b): draw an agent interaction diagram.
- 2025 Paper II, B, Q10(d) to (f) and 2024 Paper I, Q47: web hosting and publishing (10.8).
