AI<\/a> agents change that relationship. Instead of producing a single response, an agent can pursue a goal across several steps. It may open files, search for information, compare sources, update a spreadsheet, prepare a document or communicate through another application. Some agents wait for direct instructions; others monitor for a trigger and act automatically.<\/p>\nThat additional autonomy explains both the interest and the unease surrounding them. An AI assistant that drafts an email presents a limited risk. An agent with access to the inbox, customer database and shared drive can make consequential changes before anyone notices a mistake.<\/p>\n
OpenClaw<\/b> demonstrated how capable these systems can become when given broad access to a computer, persistent memory and a collection of external tools. Its open-source model also exposed the difficulties: installation remains technical, permissions can become excessive and agents may encounter malicious instructions hidden inside webpages, files or software components. Security researchers continue to regard indirect prompt injection as a fundamental vulnerability in tool-using agents.<\/p>\n
Businesses do not need to begin with that degree of autonomy. Four more accessible platforms provide very different ways to test agentic work, from a supervised desktop task to a controlled company workflow.<\/p>\n
First Understand What Makes an Agent Different<\/h2>\n
The term is used loosely. Some products described as agents are little more than customised chatbots. Others can independently select tools, decide which intermediate steps are required and execute actions in external systems.<\/p>\n
A useful definition is straightforward: an AI agent receives an objective, interprets what needs to happen, uses the tools and information available to it, and works through the task with less continuous direction than an ordinary chatbot.<\/p>\n
Three components matter.<\/p>\n
The first is the model, which interprets instructions and makes decisions. The second is context: documents, databases, previous interactions or organisational knowledge. The third is access to tools, such as a browser, email account, CRM, spreadsheet or local file system.<\/p>\n
The final component creates most of the value and most of the risk. Once the system can act, an inaccurate answer may become an incorrect database entry, an unauthorised email or an overwritten file.<\/p>\n
The appropriate agent therefore depends less on which model appears most intelligent and more on where the work happens, how repeatable it is and how much authority the organisation is prepared to delegate.<\/p>\n
For Work That Happens on the Computer: Claude Cowork<\/h2>\n
Claude Cowork<\/b> is one of the more direct demonstrations of an agent working alongside a person. Anthropic positions it as a system for multi-step knowledge work, including research synthesis, document preparation and tasks that involve files or desktop applications. It is available through the Claude desktop environment on paid plans.<\/p>\n
Cowork is relevant when the task is substantial but still irregular. It can inspect a defined collection of documents, organise material and produce an output without requiring the user to design a permanent automation.<\/p>\n
A useful first experiment would be a quarterly competitor review. Give the agent copies of five competitor reports, a folder of press releases and a previous internal briefing. Ask it to identify material changes, distinguish confirmed facts from company claims and prepare a ten-slide presentation for management.<\/p>\n
This test reveals more than asking the agent to tidy a desktop. It shows whether the system can manage several file types, maintain a consistent analytical standard and create something usable from fragmented source material.<\/p>\n
Access should remain tightly restricted. Cowork can work within designated folders, and that boundary should be treated as a minimum rather than a complete security measure. A testing folder should contain copies, not original records, and should exclude contracts, credentials, personal information and commercially sensitive documents unrelated to the task.<\/p>\n
The instruction also needs boundaries. Specify the permitted files, required output, maximum number of sources and actions that require approval. \u201cReview this folder\u201d leaves too much room for interpretation. \u201cUse only these 12 documents, do not modify the originals, identify five material developments and create a draft presentation in this output folder\u201d is considerably safer.<\/p>\n
Cowork is best understood as a supervised knowledge-work agent. It provides a visible introduction to agentic behaviour, but it should not be mistaken for an employee who can be left unattended.<\/p>\n
For Processes With Clear Rules: Zapier Agents<\/h2>\n
Many office tasks do not require an agent to roam across a computer. They follow a recognisable sequence: something happens, information is collected, a decision is made and another system is updated.<\/p>\n
Zapier Agents<\/b> is designed for this environment. It can connect an AI agent to company knowledge and thousands of applications, allowing it to work across established business tools rather than within an unrestricted desktop. Zapier currently promotes integrations across more than 9,000 apps.<\/p>\n
Consider an incoming partnership enquiry. A Zapier agent could review the submission, check whether the organisation fits predefined criteria, enrich the record with public company information, prepare a short assessment and create an opportunity in the CRM. High-potential enquiries could be routed to a senior manager; incomplete or irrelevant submissions could be placed in a separate review queue.<\/p>\n
This is a better introductory test than allowing an agent to distribute meeting tasks automatically. The organisation can define the criteria in advance and compare the agent\u2019s decisions with those of a person. It also shows where an agent improves a conventional automation. A standard workflow can move information from one field to another; an agent can interpret an unstructured message and decide which path it should follow.<\/p>\n
Zapier is relatively approachable because the process is visible as connected steps. Users can inspect which application triggers the workflow, what information is passed to the model and which action follows.<\/p>\n
That visibility does not remove the need for control. A poorly designed process can still send the wrong information to the wrong system at scale. Early workflows should draft rather than send, recommend rather than approve, and create review queues instead of making irreversible changes.<\/p>\n
Zapier suits teams that already know which recurring process they want to improve. It is less useful when the task itself is still poorly understood. Automation tends to preserve operational confusion rather than solve it.<\/p>\n
For Teams That Need Greater Technical Control: n8n<\/h2>\n
Zapier and n8n<\/b> are often grouped together because both allow users to connect applications through visual workflows. Their practical appeal is different.<\/p>\n
n8n gives technical teams more control over the workflow, underlying logic and deployment environment. Its agents can use connected applications as tools, make decisions within a workflow and combine AI with conventional rules. The platform can also be self-hosted, allowing an organisation to retain greater control over infrastructure and data flows.<\/p>\n
That makes n8n a stronger candidate when a company wants more than a quick no-code experiment but is not ready to commission a fully custom agent platform.<\/p>\n
A useful test would be an early-warning system for communications or public affairs. The workflow could collect new regulatory publications from approved sources, compare them with the organisation\u2019s policy priorities, classify their likely relevance and prepare a daily alert. Only items exceeding a defined threshold would be sent to the responsible team.<\/p>\n
This example tests several important abilities: source restriction, classification, retrieval of internal context and escalation. It also creates a natural audit trail. The team can inspect why an item was flagged and adjust the rules when the agent produces weak results.<\/p>\n
n8n requires more effort than Zapier. Non-technical users can build visual workflows, but production use benefits from someone who understands APIs, credentials, error handling and data architecture. That additional complexity is often the price of better control.<\/p>\n
It is particularly relevant for businesses operating in regulated or data-sensitive environments, provided that self-hosting is accompanied by proper security management. Hosting a system internally does not automatically make it secure. Permissions, logs, model providers and connected tools still need to be governed.<\/p>\n
For Shared Company Knowledge: Langdock<\/h2>\n
Some organisations are less interested in autonomous desktop work than in giving employees a controlled way to use AI with internal information.<\/p>\n
Langdock<\/b> approaches agents as configurable company specialists. A team can define an agent\u2019s instructions, select a model, attach approved knowledge and assign tools or actions. The agents can then be shared within a workspace with differentiated editing and usage permissions.<\/p>\n
The strongest first use case is not an all-purpose corporate assistant. It is a narrowly defined internal expert.<\/p>\n
A professional-services company could create a proposal-review agent using approved service descriptions, compliance rules, past proposals and brand guidelines. Employees would submit a draft and receive a structured review identifying unsupported claims, missing evidence, inconsistent terminology and sections requiring legal approval.<\/p>\n
This is valuable because the agent draws on a curated knowledge base rather than searching indiscriminately across the organisation. It can improve consistency without receiving unrestricted access to every company system.<\/p>\n
Langdock is aimed primarily at teams and enterprise adoption, with an emphasis on European data-protection requirements and controlled organisational use. Companies should nevertheless assess the precise configuration rather than relying on broad assurances about compliance. Relevant questions include where data is processed, which model providers receive prompts, how long content is retained and whether administrators can review agent activity.<\/p>\n
This model works best when the organisation has reliable internal material. An agent trained on obsolete policies, duplicated documents and conflicting guidance will reproduce the underlying information problem with greater confidence.<\/p>\n
For General Online Tasks: ChatGPT Agent<\/h2>\n
A fourth option is useful for users who want to experience agentic work without first building a dedicated workflow.<\/p>\n
ChatGPT<\/b> agent can navigate websites, work with uploaded files, use connected data sources, complete forms and edit spreadsheets while moving between research and action. OpenAI describes it as a system for complex online tasks that keeps the user involved in consequential steps.<\/p>\n
Its strength lies in tasks that combine research, judgement and a finite output.<\/p>\n
A relevant trial for an executive team would be preparation for a new-market meeting. Ask the agent to research a defined group of competitors, compare their positioning and pricing, extract relevant market indicators, place the findings in a spreadsheet and prepare a concise briefing. Require it to record sources and pause before accessing any connected account or submitting information.<\/p>\n
This is broader than the structured workflows suited to Zapier or n8n, but it remains more bounded than allowing a locally installed open-source agent to control the computer.<\/p>\n
ChatGPT also offers workspace agents for Business, Enterprise, Edu and Teachers plans. These are designed for shared, repeatable workflows operating within organisational permissions, although the feature remained in research preview when announced in April 2026.<\/p>\n
For individual experimentation, the general agent provides the easier starting point. For company deployment, administrators need to examine workspace permissions, connected sources and which actions require human confirmation.<\/p>\n
The First Test Should Be Deliberately Boring<\/h2>\n
The most persuasive agent demonstrations often involve ambitious assignments: conduct market research, build a presentation, contact suppliers or reorganise a project. Those examples show capability, but they are poor starting points for governance.<\/p>\n
A first test should have limited data, a measurable result and a simple recovery path. The company should already know what a good answer looks like.<\/p>\n
Suitable tasks include classifying 50 anonymised customer enquiries, checking a draft report against an approved source pack, or transferring selected fields from sample documents into a temporary spreadsheet. The agent\u2019s result can then be measured for accuracy, omissions, inappropriate actions and the amount of human correction required.<\/p>\n
The evaluation should examine more than time saved. Did the agent use information outside the permitted scope? Could reviewers reconstruct its decisions? Did it request approval at the correct point? What happened when a source contained contradictory or malicious instructions? Did the process fail safely?<\/p>\n
Prompt injection deserves particular attention. An agent researching online may encounter text instructing it to ignore its original task, disclose information or use another tool. Because language models process instructions and external content through related mechanisms, perfectly separating the two remains difficult. Anthropic\u2019s own discussion of agent containment describes security as an evolving architectural challenge rather than a solved feature.<\/p>\n
The practical response is not to avoid agents entirely. It is to limit their authority.<\/p>\n
Choose the Agent by the Boundary You Can Defend<\/h2>\n
Claude Cowork is compelling when a person wants help with a bounded collection of desktop files. Zapier is the more accessible choice for a recurring process across familiar business applications. n8n offers deeper workflow and infrastructure control. Langdock is suited to shared specialists grounded in approved company knowledge. ChatGPT agent provides a broad introduction to research-and-action tasks without requiring a workflow to be constructed first.<\/p>\n
This matters because companies often choose agents according to the most impressive demonstration. A better decision begins with permissions.<\/p>\n
Which information can the agent see? Which systems can it change? Which decisions remain with a person? How quickly can an error be detected and reversed?<\/p>\n
An agent should initially receive the minimum authority needed to prove its value. More access can be granted after the organisation understands how it behaves. Beginning with extensive permissions and trying to restrict the system later reverses the sensible order.<\/p>\n
AI agents are becoming practical business tools. Their commercial value will not come from allowing software to act everywhere. It will come from identifying the few processes where limited autonomy, reliable information and clear supervision produce a better result.<\/p>\n
<\/h1>\n
<\/p>","protected":false},"excerpt":{"rendered":"
AI agents are revolutionizing industries by enhancing efficiency, decision-making, and customer interaction. This article explores their history, current trends, and future prospects.<\/p>","protected":false},"author":5,"featured_media":13830,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[85],"tags":[],"class_list":["post-13829","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-artificial-intelligence"],"yoast_head":"\n
Four AI Agents Worth Testing - Exporis<\/title>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/www.exporis.ch\/it\/agenti-ai\/\" \/>\n<meta property=\"og:locale\" content=\"it_IT\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"Four AI Agents Worth Testing - Exporis\" \/>\n<meta property=\"og:description\" content=\"AI agents are revolutionizing industries by enhancing efficiency, decision-making, and customer interaction. This article explores their history, current trends, and future prospects.\" \/>\n<meta property=\"og:url\" content=\"https:\/\/www.exporis.ch\/it\/agenti-ai\/\" \/>\n<meta property=\"og:site_name\" content=\"Exporis\" \/>\n<meta property=\"article:published_time\" content=\"2026-07-15T04:47:15+00:00\" \/>\n<meta property=\"article:modified_time\" content=\"2026-07-15T10:12:45+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/www.exporis.ch\/wp-content\/uploads\/2026\/07\/claude-ai.jpg\" \/>\n\t<meta property=\"og:image:width\" content=\"1920\" \/>\n\t<meta property=\"og:image:height\" content=\"1072\" \/>\n\t<meta property=\"og:image:type\" content=\"image\/jpeg\" \/>\n<meta name=\"author\" content=\"Cathy\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:label1\" content=\"Scritto da\" \/>\n\t<meta name=\"twitter:data1\" content=\"Cathy\" \/>\n\t<meta name=\"twitter:label2\" content=\"Tempo di lettura stimato\" \/>\n\t<meta name=\"twitter:data2\" content=\"11 minuti\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\/\/schema.org\",\"@graph\":[{\"@type\":\"Article\",\"@id\":\"https:\/\/www.exporis.ch\/ai-agents\/#article\",\"isPartOf\":{\"@id\":\"https:\/\/www.exporis.ch\/ai-agents\/\"},\"author\":{\"name\":\"Cathy\",\"@id\":\"https:\/\/www.exporis.ch\/#\/schema\/person\/5730d61de1e8de711efd0476234e1e50\"},\"headline\":\"Four AI Agents Worth Testing\",\"datePublished\":\"2026-07-15T04:47:15+00:00\",\"dateModified\":\"2026-07-15T10:12:45+00:00\",\"mainEntityOfPage\":{\"@id\":\"https:\/\/www.exporis.ch\/ai-agents\/\"},\"wordCount\":2250,\"image\":{\"@id\":\"https:\/\/www.exporis.ch\/ai-agents\/#primaryimage\"},\"thumbnailUrl\":\"https:\/\/www.exporis.ch\/wp-content\/uploads\/2026\/07\/claude-ai.jpg\",\"articleSection\":[\"Artificial Intelligence\"],\"inLanguage\":\"it-IT\"},{\"@type\":\"WebPage\",\"@id\":\"https:\/\/www.exporis.ch\/ai-agents\/\",\"url\":\"https:\/\/www.exporis.ch\/ai-agents\/\",\"name\":\"Four AI Agents Worth Testing - Exporis\",\"isPartOf\":{\"@id\":\"https:\/\/www.exporis.ch\/#website\"},\"primaryImageOfPage\":{\"@id\":\"https:\/\/www.exporis.ch\/ai-agents\/#primaryimage\"},\"image\":{\"@id\":\"https:\/\/www.exporis.ch\/ai-agents\/#primaryimage\"},\"thumbnailUrl\":\"https:\/\/www.exporis.ch\/wp-content\/uploads\/2026\/07\/claude-ai.jpg\",\"datePublished\":\"2026-07-15T04:47:15+00:00\",\"dateModified\":\"2026-07-15T10:12:45+00:00\",\"author\":{\"@id\":\"https:\/\/www.exporis.ch\/#\/schema\/person\/5730d61de1e8de711efd0476234e1e50\"},\"breadcrumb\":{\"@id\":\"https:\/\/www.exporis.ch\/ai-agents\/#breadcrumb\"},\"inLanguage\":\"it-IT\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\/\/www.exporis.ch\/ai-agents\/\"]}]},{\"@type\":\"ImageObject\",\"inLanguage\":\"it-IT\",\"@id\":\"https:\/\/www.exporis.ch\/ai-agents\/#primaryimage\",\"url\":\"https:\/\/www.exporis.ch\/wp-content\/uploads\/2026\/07\/claude-ai.jpg\",\"contentUrl\":\"https:\/\/www.exporis.ch\/wp-content\/uploads\/2026\/07\/claude-ai.jpg\",\"width\":1920,\"height\":1072},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\/\/www.exporis.ch\/ai-agents\/#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"Home\",\"item\":\"https:\/\/www.exporis.ch\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"Four AI Agents Worth Testing\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\/\/www.exporis.ch\/#website\",\"url\":\"https:\/\/www.exporis.ch\/\",\"name\":\"Exporis\",\"description\":\"Stay Ahead Of The Game\",\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\/\/www.exporis.ch\/?s={search_term_string}\"},\"query-input\":{\"@type\":\"PropertyValueSpecification\",\"valueRequired\":true,\"valueName\":\"search_term_string\"}}],\"inLanguage\":\"it-IT\"},{\"@type\":\"Person\",\"@id\":\"https:\/\/www.exporis.ch\/#\/schema\/person\/5730d61de1e8de711efd0476234e1e50\",\"name\":\"Cathy\",\"url\":\"https:\/\/www.exporis.ch\/it\/author\/cathy\/\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"Four AI Agents Worth Testing - Exporis","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/www.exporis.ch\/it\/agenti-ai\/","og_locale":"it_IT","og_type":"article","og_title":"Four AI Agents Worth Testing - Exporis","og_description":"AI agents are revolutionizing industries by enhancing efficiency, decision-making, and customer interaction. This article explores their history, current trends, and future prospects.","og_url":"https:\/\/www.exporis.ch\/it\/agenti-ai\/","og_site_name":"Exporis","article_published_time":"2026-07-15T04:47:15+00:00","article_modified_time":"2026-07-15T10:12:45+00:00","og_image":[{"width":1920,"height":1072,"url":"https:\/\/www.exporis.ch\/wp-content\/uploads\/2026\/07\/claude-ai.jpg","type":"image\/jpeg"}],"author":"Cathy","twitter_card":"summary_large_image","twitter_misc":{"Scritto da":"Cathy","Tempo di lettura stimato":"11 minuti"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"Article","@id":"https:\/\/www.exporis.ch\/ai-agents\/#article","isPartOf":{"@id":"https:\/\/www.exporis.ch\/ai-agents\/"},"author":{"name":"Cathy","@id":"https:\/\/www.exporis.ch\/#\/schema\/person\/5730d61de1e8de711efd0476234e1e50"},"headline":"Four AI Agents Worth Testing","datePublished":"2026-07-15T04:47:15+00:00","dateModified":"2026-07-15T10:12:45+00:00","mainEntityOfPage":{"@id":"https:\/\/www.exporis.ch\/ai-agents\/"},"wordCount":2250,"image":{"@id":"https:\/\/www.exporis.ch\/ai-agents\/#primaryimage"},"thumbnailUrl":"https:\/\/www.exporis.ch\/wp-content\/uploads\/2026\/07\/claude-ai.jpg","articleSection":["Artificial Intelligence"],"inLanguage":"it-IT"},{"@type":"WebPage","@id":"https:\/\/www.exporis.ch\/ai-agents\/","url":"https:\/\/www.exporis.ch\/ai-agents\/","name":"Four AI Agents Worth Testing - Exporis","isPartOf":{"@id":"https:\/\/www.exporis.ch\/#website"},"primaryImageOfPage":{"@id":"https:\/\/www.exporis.ch\/ai-agents\/#primaryimage"},"image":{"@id":"https:\/\/www.exporis.ch\/ai-agents\/#primaryimage"},"thumbnailUrl":"https:\/\/www.exporis.ch\/wp-content\/uploads\/2026\/07\/claude-ai.jpg","datePublished":"2026-07-15T04:47:15+00:00","dateModified":"2026-07-15T10:12:45+00:00","author":{"@id":"https:\/\/www.exporis.ch\/#\/schema\/person\/5730d61de1e8de711efd0476234e1e50"},"breadcrumb":{"@id":"https:\/\/www.exporis.ch\/ai-agents\/#breadcrumb"},"inLanguage":"it-IT","potentialAction":[{"@type":"ReadAction","target":["https:\/\/www.exporis.ch\/ai-agents\/"]}]},{"@type":"ImageObject","inLanguage":"it-IT","@id":"https:\/\/www.exporis.ch\/ai-agents\/#primaryimage","url":"https:\/\/www.exporis.ch\/wp-content\/uploads\/2026\/07\/claude-ai.jpg","contentUrl":"https:\/\/www.exporis.ch\/wp-content\/uploads\/2026\/07\/claude-ai.jpg","width":1920,"height":1072},{"@type":"BreadcrumbList","@id":"https:\/\/www.exporis.ch\/ai-agents\/#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https:\/\/www.exporis.ch\/"},{"@type":"ListItem","position":2,"name":"Four AI Agents Worth Testing"}]},{"@type":"WebSite","@id":"https:\/\/www.exporis.ch\/#website","url":"https:\/\/www.exporis.ch\/","name":"Esportazione","description":"Rimani sempre un passo avanti","potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/www.exporis.ch\/?s={search_term_string}"},"query-input":{"@type":"PropertyValueSpecification","valueRequired":true,"valueName":"search_term_string"}}],"inLanguage":"it-IT"},{"@type":"Person","@id":"https:\/\/www.exporis.ch\/#\/schema\/person\/5730d61de1e8de711efd0476234e1e50","name":"Cathy","url":"https:\/\/www.exporis.ch\/it\/author\/cathy\/"}]}},"_links":{"self":[{"href":"https:\/\/www.exporis.ch\/it\/wp-json\/wp\/v2\/posts\/13829","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.exporis.ch\/it\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.exporis.ch\/it\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.exporis.ch\/it\/wp-json\/wp\/v2\/users\/5"}],"replies":[{"embeddable":true,"href":"https:\/\/www.exporis.ch\/it\/wp-json\/wp\/v2\/comments?post=13829"}],"version-history":[{"count":1,"href":"https:\/\/www.exporis.ch\/it\/wp-json\/wp\/v2\/posts\/13829\/revisions"}],"predecessor-version":[{"id":13831,"href":"https:\/\/www.exporis.ch\/it\/wp-json\/wp\/v2\/posts\/13829\/revisions\/13831"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.exporis.ch\/it\/wp-json\/wp\/v2\/media\/13830"}],"wp:attachment":[{"href":"https:\/\/www.exporis.ch\/it\/wp-json\/wp\/v2\/media?parent=13829"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.exporis.ch\/it\/wp-json\/wp\/v2\/categories?post=13829"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.exporis.ch\/it\/wp-json\/wp\/v2\/tags?post=13829"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}