<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:dc="http://purl.org/dc/elements/1.1/">
  <channel>
    <title>DEV Community: Manoj Kumar</title>
    <description>The latest articles on DEV Community by Manoj Kumar (@manoj_kumar_7bb44befb6cbb).</description>
    <link>https://dev.to/manoj_kumar_7bb44befb6cbb</link>
    <image>
      <url>https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F4232616%2F4cc6a97f-f8ab-498b-a986-2b8ea6929b61.jpg</url>
      <title>DEV Community: Manoj Kumar</title>
      <link>https://dev.to/manoj_kumar_7bb44befb6cbb</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/manoj_kumar_7bb44befb6cbb"/>
    <language>en</language>
    <item>
      <title>I Connected Two AI Agents to a Newsroom with Switch. Here's What Actually Happened.</title>
      <dc:creator>Manoj Kumar</dc:creator>
      <pubDate>Sat, 07 Nov 2026 18:16:18 +0000</pubDate>
      <link>https://dev.to/manoj_kumar_7bb44befb6cbb/i-connected-two-ai-agents-to-a-newsroom-with-switch-heres-what-actually-happened-4442</link>
      <guid>https://dev.to/manoj_kumar_7bb44befb6cbb/i-connected-two-ai-agents-to-a-newsroom-with-switch-heres-what-actually-happened-4442</guid>
      <description>&lt;h1&gt;
  
  
  I Connected Two AI Agents to a Newsroom with Switch. Here's What Actually Happened.
&lt;/h1&gt;

&lt;p&gt;How I built a Researcher–Skeptic workflow, debugged a missing agent response, and learned to make AI collaboration more reliable.&lt;/p&gt;

&lt;p&gt;By Manoj Kumar Dadala | Flip the Switch Hackathon 2026&lt;/p&gt;

&lt;h2&gt;
  
  
  From talking to AI agents to working with them
&lt;/h2&gt;

&lt;p&gt;Most of my experience with AI agents began in individual coding sessions.&lt;/p&gt;

&lt;p&gt;I would give an agent a task, wait for its response, review the output, and decide what to do next.&lt;/p&gt;

&lt;p&gt;That works when one person is controlling one agent.&lt;/p&gt;

&lt;p&gt;But real teams work differently.&lt;/p&gt;

&lt;p&gt;A task appears in a shared conversation. Different people contribute information, challenge assumptions, and make decisions. The relevant context needs to remain visible to everyone.&lt;/p&gt;

&lt;p&gt;I wanted to explore what happens when AI agents join that process.&lt;/p&gt;

&lt;p&gt;That's what drew me to Switch.&lt;/p&gt;

&lt;p&gt;Switch connects AI agents to team-chat applications so they can participate in shared rooms alongside people.&lt;/p&gt;

&lt;p&gt;For the Flip the Switch Hackathon, I used it to experiment with an AI-assisted newsroom called TruthDesk.&lt;/p&gt;

&lt;p&gt;The goal was simple:&lt;/p&gt;

&lt;p&gt;One human editor, two AI agents, one shared investigation.&lt;/p&gt;

&lt;p&gt;What followed taught me some unexpected lessons about AI-agent orchestration.&lt;/p&gt;

&lt;h2&gt;
  
  
  1. Understanding the Switch workflow
&lt;/h2&gt;

&lt;p&gt;Switch brings agents into a collaborative environment.&lt;/p&gt;

&lt;p&gt;Instead of interacting with an agent only through its own terminal or application, a team can communicate through an existing chat platform.&lt;/p&gt;

&lt;p&gt;Switch provides the connection between the conversation and the agent.&lt;/p&gt;

&lt;p&gt;For my project, the components were:&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Component&lt;/th&gt;
&lt;th&gt;Responsibility&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Mattermost&lt;/td&gt;
&lt;td&gt;Shared newsroom conversation&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Switch&lt;/td&gt;
&lt;td&gt;Room association, message ingestion, and agent routing&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Researcher&lt;/td&gt;
&lt;td&gt;Investigates claims and identifies evidence&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Skeptic&lt;/td&gt;
&lt;td&gt;Challenges interpretations and missing context&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Human editor&lt;/td&gt;
&lt;td&gt;Reviews findings and makes decisions&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;TruthDesk&lt;/td&gt;
&lt;td&gt;Independently verifies quotations and prepares evidence reports&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;One distinction is important: &lt;strong&gt;Switch is the collaboration layer; TruthDesk is the evidence-verification application I built around that collaboration.&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;They solve different problems.&lt;/p&gt;

&lt;h2&gt;
  
  
  2. Setting up my first newsroom
&lt;/h2&gt;

&lt;p&gt;I worked with a local Switch environment and Mattermost.&lt;/p&gt;

&lt;p&gt;I created a newsroom channel called:&lt;/p&gt;

&lt;p&gt;&lt;code&gt;truthdesk-newsroom&lt;/code&gt;&lt;/p&gt;

&lt;p&gt;The next step was associating the Mattermost channel with a Switch room and making the agents available to that room.&lt;/p&gt;

&lt;p&gt;I used two separate agent identities:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;&lt;code&gt;truthdesk-researcher&lt;/code&gt;&lt;/li&gt;
&lt;li&gt;&lt;code&gt;truthdesk-skeptic&lt;/code&gt;&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Why two agents?&lt;/p&gt;

&lt;p&gt;Because I didn't want two assistants producing similar summaries.&lt;/p&gt;

&lt;p&gt;I wanted two different responsibilities.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Researcher:&lt;/strong&gt; Examine the claim and identify relevant source material.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Skeptic:&lt;/strong&gt; Challenge the Researcher's interpretation and identify conclusions that go beyond the source.&lt;/p&gt;

&lt;p&gt;The intended conversation was:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;Human editor
    |
    | Posts an investigation request
    v
Mattermost newsroom
    |
    | Connected through Switch
    v
Researcher + Skeptic
    |
    | Return their findings
    v
Human reviews the discussion
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;The room provides the shared context for this collaboration.&lt;/p&gt;

&lt;p&gt;It also makes the interaction inspectable: the human can see what was requested and what each agent actually said.&lt;/p&gt;

&lt;h3&gt;
  
  
  A useful lesson for beginners
&lt;/h3&gt;

&lt;p&gt;Before configuring multiple agents, verify one simple interaction.&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;Create or select your chat channel.&lt;/li&gt;
&lt;li&gt;Associate it with a Switch room.&lt;/li&gt;
&lt;li&gt;Make one agent available in that room.&lt;/li&gt;
&lt;li&gt;Send a short request addressed to that agent.&lt;/li&gt;
&lt;li&gt;Confirm an actual response appears.&lt;/li&gt;
&lt;li&gt;Only then introduce a second agent.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;Don't start by debugging a complex multi-agent task.&lt;/p&gt;

&lt;p&gt;First prove that the basic message path works.&lt;/p&gt;

&lt;p&gt;For installation and configuration, start with the &lt;a href="https://www.wemakedevs.org/hackathons/switch/resources" rel="noopener noreferrer"&gt;official Switch resources&lt;/a&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  3. Giving the agents a real task
&lt;/h2&gt;

&lt;p&gt;I chose a newsroom investigation because it naturally requires multiple perspectives.&lt;/p&gt;

&lt;p&gt;For testing, I used a fictional city-council notice involving electric scooters.&lt;/p&gt;

&lt;p&gt;The claim suggested that a city had banned all electric scooters.&lt;/p&gt;

&lt;p&gt;The synthetic source document actually described something narrower: rental scooters were restricted on one street during particular hours.&lt;/p&gt;

&lt;p&gt;That difference made the investigation useful.&lt;/p&gt;

&lt;p&gt;The Researcher's responsibility was to identify what the source supported.&lt;/p&gt;

&lt;p&gt;The Skeptic's responsibility was to challenge an interpretation that extended the restriction to every scooter across the city.&lt;/p&gt;

&lt;p&gt;During earlier testing, I observed authentic Researcher and Skeptic activity in the Mattermost newsroom.&lt;/p&gt;

&lt;p&gt;This was the first meaningful milestone.&lt;/p&gt;

&lt;p&gt;The agents were participating in a shared environment rather than producing isolated answers in separate sessions.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;[Screenshot: Historical Researcher and Skeptic responses in Mattermost]&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;But observing agent responses also raised another question:&lt;/p&gt;

&lt;p&gt;How could I verify the evidence they provided?&lt;/p&gt;

&lt;h2&gt;
  
  
  4. Adding an independent evidence checker
&lt;/h2&gt;

&lt;p&gt;This is where TruthDesk came in.&lt;/p&gt;

&lt;p&gt;I built a Python Evidence Engine that compares agent quotations with separately registered source documents.&lt;/p&gt;

&lt;p&gt;For each finding, the system records:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Investigation ID&lt;/li&gt;
&lt;li&gt;Agent identity&lt;/li&gt;
&lt;li&gt;Source ID&lt;/li&gt;
&lt;li&gt;Exact quotation&lt;/li&gt;
&lt;li&gt;Agent conclusion&lt;/li&gt;
&lt;li&gt;Citation-verification result&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The important design choice was keeping &lt;strong&gt;agent conclusions separate from source verification&lt;/strong&gt;.&lt;/p&gt;

&lt;p&gt;Consider the synthetic scooter notice.&lt;/p&gt;

&lt;p&gt;The original text refers to:&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;rental electric scooters&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;Now deliberately change the quotation to:&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;all electric scooters&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;The meaning changes.&lt;/p&gt;

&lt;p&gt;My Evidence Engine detects that the altered quotation is not present in the registered source.&lt;/p&gt;

&lt;p&gt;The original quotation returns &lt;code&gt;VERIFIED&lt;/code&gt;.&lt;/p&gt;

&lt;p&gt;The modified quotation returns &lt;code&gt;FAILED&lt;/code&gt;.&lt;/p&gt;

&lt;p&gt;This demonstration uses deterministic synthetic fixtures. It does not imply that the historical agents fabricated those words.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;[Screenshot: TruthDesk VERIFIED versus FAILED comparison]&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;The larger lesson is applicable to other Switch projects.&lt;/p&gt;

&lt;p&gt;If your agents produce information that matters, consider what evidence can independently confirm their outputs.&lt;/p&gt;

&lt;p&gt;An agent's response is a record of what it said.&lt;/p&gt;

&lt;p&gt;It isn't automatically proof that what it said was correct.&lt;/p&gt;

&lt;h2&gt;
  
  
  5. The debugging incident that changed my understanding
&lt;/h2&gt;

&lt;p&gt;After observing historical agent activity, I tried another Mattermost connectivity test.&lt;/p&gt;

&lt;p&gt;The message appeared in the chat channel.&lt;/p&gt;

&lt;p&gt;But the expected agent response never arrived.&lt;/p&gt;

&lt;p&gt;Initially, this looked like an agent problem.&lt;/p&gt;

&lt;p&gt;Perhaps the model had reached a usage limit.&lt;/p&gt;

&lt;p&gt;Perhaps the agent instructions were wrong.&lt;/p&gt;

&lt;p&gt;Perhaps the mention wasn't recognized.&lt;/p&gt;

&lt;p&gt;Instead of guessing, I examined the available message and bridge evidence.&lt;/p&gt;

&lt;p&gt;I confirmed that:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;The Mattermost post existed.&lt;/li&gt;
&lt;li&gt;Its channel-to-room association was valid.&lt;/li&gt;
&lt;li&gt;The relevant agent was still a room member.&lt;/li&gt;
&lt;li&gt;No corresponding Switch message mapping existed for the target post.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Then I found an important log entry:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;AgentConnection sweep skipped:
the event loop was blocked for 3712.4s
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Switch subsequently reported that its lease had not been renewed in time.&lt;/p&gt;

&lt;p&gt;The controller connection closed, the Mattermost delivery consumer stopped, and websocket connections disconnected before the bridge restarted.&lt;/p&gt;

&lt;p&gt;That evidence indicated that the target request had failed at the &lt;strong&gt;ingestion stage&lt;/strong&gt;.&lt;/p&gt;

&lt;p&gt;The agent wasn't merely refusing to answer.&lt;/p&gt;

&lt;p&gt;There was no evidence that it had been invoked for that post.&lt;/p&gt;

&lt;p&gt;I also found a separate Researcher usage-limit event, but it belonged to a different message.&lt;/p&gt;

&lt;p&gt;That distinction matters.&lt;/p&gt;

&lt;p&gt;Blaming the target failure on quota would have been an unsupported conclusion.&lt;/p&gt;

&lt;h3&gt;
  
  
  What this taught me
&lt;/h3&gt;

&lt;p&gt;A message appearing in Mattermost proves that Mattermost stored it.&lt;/p&gt;

&lt;p&gt;It does not prove that Switch imported it.&lt;/p&gt;

&lt;p&gt;A message existing inside Switch does not necessarily prove an agent executed.&lt;/p&gt;

&lt;p&gt;An agent producing output does not prove the output is correct.&lt;/p&gt;

&lt;p&gt;These are separate stages requiring separate checks.&lt;/p&gt;

&lt;h2&gt;
  
  
  6. A troubleshooting checklist for new Switch builders
&lt;/h2&gt;

&lt;p&gt;When an agent doesn't respond, investigate in this order.&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Check&lt;/th&gt;
&lt;th&gt;Question&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Chat platform&lt;/td&gt;
&lt;td&gt;Was the message actually posted?&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Channel association&lt;/td&gt;
&lt;td&gt;Is the intended channel connected to the expected Switch room?&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Room membership&lt;/td&gt;
&lt;td&gt;Is the target agent available in that room?&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Ingestion&lt;/td&gt;
&lt;td&gt;Did Switch receive the post?&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Addressing&lt;/td&gt;
&lt;td&gt;Was the intended agent identified?&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Execution&lt;/td&gt;
&lt;td&gt;Did an agent invocation occur?&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Provider&lt;/td&gt;
&lt;td&gt;Was execution blocked by authentication, connectivity, or usage limits?&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Response&lt;/td&gt;
&lt;td&gt;Was the output delivered back to the conversation?&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;The key is locating the first stage without evidence of success.&lt;/p&gt;

&lt;p&gt;For example:&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;If Mattermost stored the message but Switch never ingested it&lt;/strong&gt;, changing the agent prompt is unlikely to solve the problem.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;If Switch routed the message but the model provider rejected execution&lt;/strong&gt;, debugging the bridge is unlikely to solve that problem.&lt;/p&gt;

&lt;p&gt;Treat each layer independently.&lt;/p&gt;

&lt;p&gt;That's the most practical lesson I learned during this hackathon.&lt;/p&gt;

&lt;h2&gt;
  
  
  7. Designing for agent failures
&lt;/h2&gt;

&lt;p&gt;The experience changed TruthDesk's workflow design.&lt;/p&gt;

&lt;p&gt;Instead of displaying an optimistic status such as “Agent completed,” I added local evidence states:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Not Captured&lt;/li&gt;
&lt;li&gt;Captured Pending Import&lt;/li&gt;
&lt;li&gt;Imported/Validated&lt;/li&gt;
&lt;li&gt;Rejected&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;These statuses represent observed local actions.&lt;/p&gt;

&lt;p&gt;They do not claim that Switch successfully delivered a request or executed an agent.&lt;/p&gt;

&lt;p&gt;I also preserved original agent messages, source references, and content hashes.&lt;/p&gt;

&lt;p&gt;Completed imports are protected against overwrite.&lt;/p&gt;

&lt;p&gt;For future automation, I implemented an offline-tested Mattermost read-only connector foundation using documented post-retrieval APIs and explicit author identity mappings.&lt;/p&gt;

&lt;p&gt;It is not yet connected to an authorized live Mattermost identity.&lt;/p&gt;

&lt;p&gt;That limitation is intentional: I didn't want a demonstration to depend on extracting credentials or bypassing administrative permissions.&lt;/p&gt;

&lt;h2&gt;
  
  
  8. What I would build differently next time
&lt;/h2&gt;

&lt;p&gt;If I restarted this project, I would organize development around four milestones.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Milestone 1: One reliable agent&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;Establish one verified request-and-response interaction.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Milestone 2: Two distinct agents&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;Give the second agent a clear responsibility that improves the human's decision.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Milestone 3: Observable execution&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;Record which messages were posted, ingested, routed, executed, and answered.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Milestone 4: Independently checked output&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;Verify important claims or artifacts against evidence the agents don't control.&lt;/p&gt;

&lt;p&gt;I would also keep the human approval boundary explicit from the beginning.&lt;/p&gt;

&lt;p&gt;For TruthDesk, publication remains refused unless the separate approval requirements are satisfied.&lt;/p&gt;

&lt;p&gt;The evidence engine deliberately does not issue a real-world factual truth verdict.&lt;/p&gt;

&lt;h2&gt;
  
  
  9. What works today
&lt;/h2&gt;

&lt;p&gt;TruthDesk currently has:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Historical Switch-mediated Mattermost agent activity.&lt;/li&gt;
&lt;li&gt;A local Researcher/Skeptic investigation workflow.&lt;/li&gt;
&lt;li&gt;Deterministic citation verification.&lt;/li&gt;
&lt;li&gt;Source provenance and SHA-256 hashes.&lt;/li&gt;
&lt;li&gt;Agent conclusion and conflict reporting.&lt;/li&gt;
&lt;li&gt;Persistent local workflow states.&lt;/li&gt;
&lt;li&gt;A synthetic VERIFIED/FAILED demonstration.&lt;/li&gt;
&lt;li&gt;Human approval and publication-refusal safeguards.&lt;/li&gt;
&lt;li&gt;An offline-tested Mattermost connector foundation.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The local Python tests passed at the latest reported development checkpoint.&lt;/p&gt;

&lt;p&gt;However, the project does not yet have reliable fully automated end-to-end Switch delivery and TruthDesk import.&lt;/p&gt;

&lt;p&gt;That is a real limitation of the current prototype.&lt;/p&gt;

&lt;h2&gt;
  
  
  10. Explore the project
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;Source code:&lt;/strong&gt; &lt;a href="https://github.com/Roy0x/truthdesk" rel="noopener noreferrer"&gt;TruthDesk on GitHub&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;To run the local dashboard after following the repository's installation instructions:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight powershell"&gt;&lt;code&gt;&lt;span class="n"&gt;python&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="nt"&gt;-m&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="nx"&gt;truthdesk_evidence.dashboard_cli&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="se"&gt;`
&lt;/span&gt;&lt;span class="w"&gt;  &lt;/span&gt;&lt;span class="nt"&gt;--host&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="nx"&gt;127.0.0.1&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="se"&gt;`
&lt;/span&gt;&lt;span class="w"&gt;  &lt;/span&gt;&lt;span class="nt"&gt;--port&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="nx"&gt;8765&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="se"&gt;`
&lt;/span&gt;&lt;span class="w"&gt;  &lt;/span&gt;&lt;span class="nt"&gt;--work-dir&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;&lt;span class="nx"&gt;work\investigations&lt;/span&gt;&lt;span class="w"&gt;
&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Open:&lt;/p&gt;

&lt;p&gt;&lt;code&gt;http://127.0.0.1:8765/&lt;/code&gt;&lt;/p&gt;

&lt;p&gt;The synthetic evidence demonstration is available at:&lt;/p&gt;

&lt;p&gt;&lt;code&gt;http://127.0.0.1:8765/demo/proof&lt;/code&gt;&lt;/p&gt;

&lt;p&gt;For your own Switch setup, begin with the official &lt;a href="https://www.wemakedevs.org/hackathons/switch/resources" rel="noopener noreferrer"&gt;Switch quickstart and resources&lt;/a&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  11. The biggest lesson
&lt;/h2&gt;

&lt;p&gt;Before this hackathon, I mostly thought of agent reliability as a question of whether the AI could complete its assigned task.&lt;/p&gt;

&lt;p&gt;Building with Switch taught me that reliability starts earlier.&lt;/p&gt;

&lt;p&gt;Did the message reach the room?&lt;/p&gt;

&lt;p&gt;Did the correct agent receive it?&lt;/p&gt;

&lt;p&gt;Did execution happen?&lt;/p&gt;

&lt;p&gt;Did the response arrive?&lt;/p&gt;

&lt;p&gt;And can we independently verify what the response says?&lt;/p&gt;

&lt;p&gt;A team needs answers to all of those questions before it can rely on an AI workflow.&lt;/p&gt;

&lt;p&gt;That's what I learned while building TruthDesk.&lt;/p&gt;

&lt;p&gt;I still have integration work to finish, but I've come away with a clearer understanding of what trustworthy agent collaboration requires.&lt;/p&gt;

&lt;p&gt;A good agent workflow doesn't just produce an answer. It makes the path to that answer observable.&lt;/p&gt;




&lt;p&gt;Author:Manoj Kumar Dadala&lt;br&gt;&lt;br&gt;
Project:&lt;a href="https://github.com/Roy0x/truthdesk" rel="noopener noreferrer"&gt;TruthDesk&lt;/a&gt;&lt;br&gt;&lt;br&gt;
Event:&lt;a href="https://www.wemakedevs.org/hackathons/switch" rel="noopener noreferrer"&gt;Flip the Switch Hackathon 2026&lt;/a&gt;&lt;br&gt;&lt;br&gt;
Tags:#Switch #AI #Agents #BuildInPublic&lt;/p&gt;

</description>
      <category>ai</category>
    </item>
  </channel>
</rss>
