[{"data":1,"prerenderedAt":583},["ShallowReactive",2],{"footer-vacatures-en":3,"case-robo-helma-en":4,"content-by-keys-useCases-en-ai-agent-laten-bouwen,interne-ai-assistent-laten-ontwikkelen,eigen-data-en-kennis-gebruiken-met-ai,ai-toevoegen-aan-bestaande-software,tests-evals-monitoring-voor-ai":186},[],{"id":5,"title":6,"approach":7,"body":22,"build":29,"card":42,"connections":46,"cta":47,"dataExamples":46,"description":26,"extension":53,"facts":54,"featured":69,"guardrails":70,"hero":83,"integrations":46,"key":88,"meta":89,"navigation":69,"offers":90,"order":100,"path":101,"proof":102,"relations":115,"result":136,"seo":148,"situation":151,"stem":164,"summary":165,"usecases":166,"visuals":178,"__hash__":185},"cases_en\u002Fcases\u002Frobo-helma.md","Robo Helma: AI support that knows what it is talking about",{"title":8,"steps":9},"How did we approach it?",[10,13,16,19],{"title":11,"text":12},"Split knowledge by function","We treat behaviour and support rules as OpenAI Skills, product documentation as searchable knowledge in a vector store and current customer or system information as external context. This gives every kind of information a clear place.\n",{"title":14,"text":15},"Retrieve context before answering","For each support question, Helma first gathers the relevant context. Product knowledge comes via File Search, customer information can be retrieved through an external MCP connection and current incident information comes from a separate controlled source.\n",{"title":17,"text":18},"Connect engineering and support","A support ticket can be linked to an existing GitHub issue. Helma follows the technical status read-only. When engineering closes or reopens an issue, the new status can be included in a customer-safe draft answer without exposing internal development information.\n",{"title":20,"text":21},"People keep the final say","Helma drafts an answer for human review. The support agent can send it directly or edit it. Both the original AI answer and the answer that was eventually sent are stored, so we can measure how useful the AI really is.\n",{"type":23,"value":24,"toc":25},"minimark",[],{"title":26,"searchDepth":27,"depth":27,"links":28},"",2,[],{"title":30,"intro":31,"points":32},"What did we build?","Robo Helma is a complete support workflow in which AI, knowledge, integrations, ticketing and feedback come together, not a standalone chatbot next to an existing helpdesk.\n",[33,36,39],{"title":34,"text":35},"A configurable knowledge layer","Every brand can use its own OpenAI Skills and its own vector store. Skills contain, for example, support behaviour, tone of voice and escalation policy. The vector store holds product knowledge such as FAQs, policies, known issues and documentation.\n",{"title":37,"text":38},"Context from real systems","Helma can combine product knowledge with customer-specific information, current incidents, previous customer interactions and linked engineering context. That way the model does not have to make up critical facts itself.\n",{"title":40,"text":41},"A complete support environment","Around the AI we built a multi-tenant inbox, ticket statuses, internal notes, attachments, reporting, feedback tracking and integrations with, among others, email, Facebook Messenger, app store reviews and GitHub.\n",{"tag":43,"title":6,"summary":44,"metric":45},"AI support · human-in-the-loop","Robo Helma combines product documentation, support rules, customer data, incident information and GitHub status in one controlled AI workflow.","67% of generated replies are sent without edits.",null,{"title":48,"body":49,"primaryCta":50,"secondaryCta":51,"local":52},"Building AI that has to do more than generate neat text?","Show us which knowledge and systems are involved in your process. We help you work out how to turn that into a reliable AI workflow, including integrations, guardrails and human-in-the-loop.\n","Book a call","View offer","Senior AI and software engineering from Delft, for organisations in Delft, The Hague, Rotterdam and the surrounding area.\n","md",{"title":55,"items":56},"The case in brief",[57,60,63,66],{"label":58,"value":59},"Ready to send","67% sent without edits",{"label":61,"value":62},"Small edit","18%",{"label":64,"value":65},"Fully rewritten","15%",{"label":67,"value":68},"Knowledge","Skills, vector search and live context",true,{"title":71,"intro":72,"points":73},"Reliable AI takes more than good prompts","Customers can put anything in a message and external sources can contain outdated or unreliable information. That is why we do not automatically treat that input as an instruction for the model.\n",[74,77,80],{"title":75,"text":76},"Prompt injection is checked explicitly","Helma checks customer messages for attempts to manipulate instructions, safety rules, tools or hidden prompts.\n",{"title":78,"text":79},"Every source has its own role","Product documentation determines product facts, customer data determines customer-specific information and GitHub only provides technical status. One source cannot simply override another.\n",{"title":81,"text":82},"No automatic promises from engineering data","A closed GitHub issue, for example, does not automatically mean that a fix has been deployed or is available. That boundary is explicitly built into the AI workflow.\n",{"headline":84,"intro":85,"primaryCta":86,"secondaryCta":87},"Two out of three AI answers are good enough to send right away.","A language model can write a friendly answer in seconds. The hard part is making sure the answer is correct: with the right product knowledge, support rules, current customer information and the status of technical issues. Robo Helma brings those sources together and drafts an answer that a support agent reviews before it goes to the customer.\n","See how Helma works","Tell us what you want to build","robo-helma",{},{"eyebrow":91,"primary":92},"Matching offer",{"key":93,"title":94,"text":95,"price":96,"result":97,"duration":98,"nextStep":99},"build-integrate","Build & Integrate","Do you have an AI challenge of your own where multiple knowledge sources, existing systems and reliable AI need to come together? Then we design and build the workflow and integrations around your situation.\n","Fixed price after scoping","Production-ready AI integrated into your existing environment","Depends on scope","First we determine which knowledge, systems and controls are really needed",0,"\u002Fcases\u002Frobo-helma",{"title":103,"intro":104,"points":105},"Proof","The reliability comes from the engineering around the model. Robo Helma stores the context used for every AI run and measures what support agents eventually do with the generated answer.\n",[106,109,112],{"title":107,"text":108},"67% direct AI acceptance","Two out of three generated answers are currently sent without substantive edits. Another 18% is only slightly adjusted.\n",{"title":110,"text":111},"Source information is visible","In the AI analysis, a support agent can see which knowledge references were used via File Search, including the files used and available relevance scores.\n",{"title":113,"text":114},"Improvement is part of the system","Reporting compares the generated draft with the answer that was actually sent and classifies it as accepted, edited or rejected. That way we can measure whether the AI is getting better and where improvement is needed.\n",{"buyerTriggerKeys":116,"useCaseKeys":118,"offerKeys":124,"solutionKey":88,"capabilityKeys":125},[117],"ai-laten-bouwen",[119,120,121,122,123],"ai-agent-laten-bouwen","interne-ai-assistent-laten-ontwikkelen","eigen-data-en-kennis-gebruiken-met-ai","ai-toevoegen-aan-bestaande-software","tests-evals-monitoring-voor-ai",[93],[126,127,128,129,130,131,132,133,134,135],"ai-engineering","software-engineering","retrieval","vector-search","ai-agents","integrations","human-in-the-loop","evaluation","monitoring","security",{"title":137,"items":138},"What does it deliver?",[139,142,145],{"title":140,"text":141},"67% can go straight to the customer","Currently, 67% of the support answers generated by Robo Helma are sent without edits. The support agent only needs to review and send the answer.\n",{"title":143,"text":144},"Another 18% only needs an edit","For 18%, the generated answer is largely usable, but a support agent adjusts something before it is sent.\n",{"title":146,"text":147},"The remaining 15% shows where we can improve further","About 15% of the answers are fully rewritten. Because we store the original AI draft and the final answer, we can see exactly where Helma still falls short and keep improving on that.\n",{"title":149,"description":150},"Robo Helma: AI helpdesk with knowledge and integrations | 9to5 software","Case: how 9to5 software built Robo Helma, an AI helpdesk that combines product documentation, skills, customer data and GitHub. 67% of the AI answers are currently sent without edits.\n",{"title":152,"intro":153,"points":154},"Where was the real problem?","Automating support looks like a prompt problem at first sight. In practice, the difficulty lies elsewhere. A good answer requires different kinds of knowledge, each with its own source, freshness and reliability.\n",[155,158,161],{"title":156,"text":157},"Product knowledge changes constantly","FAQs, manuals, policies, known issues and other documentation must be available to the AI without stuffing everything into one huge prompt or hard-coding knowledge by hand.\n",{"title":159,"text":160},"Not all knowledge is documentation","Tone of voice, support policy, escalation rules and ways of working are instructions. Customer data and current incidents are live data. You cannot reliably treat these different kinds of context as if they were all the same.\n",{"title":162,"text":163},"A convincing answer can still be wrong","A model can neatly make up something that sounds plausible. For support, that is not enough. The AI must know which sources are leading, when information is missing and when a person must review or take over.\n","cases\u002Frobo-helma","Robo Helma is an AI-first helpdesk that helps support agents give good answers faster. Instead of just putting a language model on an inbox, we combine product documentation, support rules, customer data, incident information and GitHub status into one controlled AI workflow. Currently, 67% of the generated answers are sent to the customer without edits.\n",{"title":167,"items":168},"Where is this relevant?",[169,172,175],{"title":170,"text":171},"Our support agents keep gathering the same information","Product knowledge, customer data and technical status are spread over different systems before anyone can give a good answer.\n",{"title":173,"text":174},"We want AI support, but not a hallucinating chatbot","Answers must be based on verifiable knowledge and current data, with clear limits on what the AI may and may not assume.\n",{"title":176,"text":177},"We want to know whether the AI is actually getting better","Not just counting generated answers, but measuring how many drafts are usable without changes and where support agents still need to correct them.\n",{"hero":179,"situation":180,"approach":181,"build":182,"proof":183,"teaser":184},"\u002Fmedia\u002Fscreenshots\u002Fhelma-hero.webm","\u002Fmedia\u002Fscreenshots\u002Fhelma-knowledge.webp","ls-workflow-flow","\u002Fmedia\u002Fscreenshots\u002Fhelma-analysis.webp","\u002Fmedia\u002Fscreenshots\u002Fhelma-reporting.webp","\u002Fmedia\u002Fscreenshots\u002Fhelma-teaser.webm","7y61CkSiqDr7SMbpk_FIqfwfxBE2egC3_6tKSFy5DHQ",[187,269,349,425,502],{"id":188,"title":189,"body":190,"description":26,"desiredOutcome":194,"extension":53,"hero":206,"key":119,"meta":208,"navigation":69,"path":209,"problem":210,"proof":222,"recognition":234,"recommendedApproach":249,"relations":251,"seo":259,"stem":262,"summary":263,"visuals":264,"__hash__":268},"useCases_en\u002Fuse-cases\u002Fai-agent-laten-bouwen.md","Have an AI agent built",{"type":23,"value":191,"toc":192},[],{"title":26,"searchDepth":27,"depth":27,"links":193},[],{"intro":195,"items":196},"You don't want a standalone AI demo, but an agent that performs a clear task and fits the software and processes that already exist.\n",[197,200,203],{"title":198,"text":199},"A working end-to-end flow","The agent can retrieve the relevant information, make decisions and take the right actions.\n",{"title":201,"text":202},"Control over what it may do","Permissions, validation and human oversight limit its autonomy where needed.\n",{"title":204,"text":205},"Manageable in production","Errors, model behaviour, costs and tool calls are visible enough to use the agent responsibly.\n",{"intro":207},"An agent that only gives clever answers won't do much for you. It becomes useful when it can retrieve information, make choices and safely take actions in your existing systems. We build that whole technical setup.\n",{},"\u002Fuse-cases\u002Fai-agent-laten-bouwen",{"intro":211,"causes":212},"An AI agent is more than a prompt with a few functions. It is a software system in which unpredictable models are allowed to take real actions.\n",[213,216,219],{"title":214,"text":215},"Autonomy introduces risk","A wrong model decision can have consequences outside the chat interface.\n",{"title":217,"text":218},"Integrations determine the real value","Without good access to existing data and systems, the agent remains mostly a demonstration.\n",{"title":220,"text":221},"Reliability is built around the model","Validation, retries, state, logging and human-in-the-loop are often just as important as the choice of model.\n",{"intro":223,"points":224},"9to5 software combines AI engineering with senior software engineering. So next to the smart part, we also build the technical environment you need to use an agent seriously.\n",[225,228,231],{"title":226,"text":227},"Agents and regular software","Tooling, APIs, data, auth and error handling belong to the same solution.\n",{"title":229,"text":230},"Human-in-the-loop where needed","Not every process gets better with maximum autonomy.\n",{"title":232,"text":233},"No black box left behind","We document the important choices so the system stays transferable.\n",{"intro":235,"signals":236},"The use case is often already clear. The technical challenge is running the complete workflow reliably.\n",[237,240,243,246],{"title":238,"text":239},"The agent has to do more than answer","It has to retrieve data, call systems or actually change something.\n",{"title":241,"text":242},"Several tools or systems are involved","The workflow runs across APIs, databases or existing business software.\n",{"title":244,"text":245},"Not every action may run automatically","For critical steps a user must be able to check or approve.\n",{"title":247,"text":248},"A simple prompt is not enough","State, context, error handling and tool use become part of the technical design.\n",{"intro":250},"With Build & Integrate we design and build the complete agent flow: architecture, tools, integrations, permissions, guardrails and production behaviour. When a new or uncertain agent concept first needs to be proven technically, an AI Prototype Sprint can be a more logical first step.\n",{"buyerTriggerKeys":252,"primaryOfferKey":93,"offerKeys":253,"caseKeys":255,"relatedUseCaseKeys":256},[117],[93,254],"ai-prototype-sprint",[],[257,258],"agentic-workflows-bestaande-software","bedrijfsproces-automatiseren-met-ai",{"title":260,"description":261},"AI agent development (Claude, ChatGPT, MCP) | 9to5 software","Need an AI agent for a serious business process? 9to5 builds agents with Claude or OpenAI, MCP tools and integrations with your existing software and data.\n","use-cases\u002Fai-agent-laten-bouwen","You know which task or workflow the agent has to perform. Now it needs to be designed well, connected and made reliable enough for real use.\n",{"hero":265,"problem":266,"proof":267},"sq-process-steps","sq-guardrails-check","ls-agentic-loop","Sz6OZV7bqm5zl00ThdeWnUL9u1nZHw5GFlz7cgVRcnE",{"id":270,"title":271,"body":272,"description":26,"desiredOutcome":276,"extension":53,"hero":288,"key":120,"meta":290,"navigation":69,"path":291,"problem":292,"proof":304,"recognition":316,"recommendedApproach":331,"relations":333,"seo":339,"stem":342,"summary":343,"visuals":344,"__hash__":348},"useCases_en\u002Fuse-cases\u002Finterne-ai-assistent-laten-ontwikkelen.md","Have an internal AI assistant developed",{"type":23,"value":273,"toc":274},[],{"title":26,"searchDepth":27,"depth":27,"links":275},[],{"intro":277,"items":278},"Employees should get to the right information faster without creating a new knowledge problem with unreliable answers.\n",[279,282,285],{"title":280,"text":281},"Answers based on your own sources","The assistant uses relevant internal documents, data and knowledge.\n",{"title":283,"text":284},"Sources stay verifiable","Users can see where important information comes from.\n",{"title":286,"text":287},"Access follows existing permissions","The assistant doesn't reveal information the user normally isn't allowed to see.\n",{"intro":289},"An internal assistant only becomes truly useful when it knows your context, can find sources and only shows information a user is allowed to see. We build the technology behind that experience.\n",{},"\u002Fuse-cases\u002Finterne-ai-assistent-laten-ontwikkelen",{"intro":293,"causes":294},"An internal AI assistant is mainly an information and integration challenge around a language model.\n",[295,298,301],{"title":296,"text":297},"Knowledge is scattered","Documents, databases and other systems each have their own structure.\n",{"title":299,"text":300},"A model doesn't know what is true","Without good retrieval and source selection, convincing-sounding nonsense can appear.\n",{"title":302,"text":303},"Permissions have to carry through","Central access to knowledge must not bypass existing security boundaries.\n",{"intro":305,"points":306},"The chat window is the easy part. Whether employees can rely on the assistant depends on how data, permissions, retrieval and AI are connected.\n",[307,310,313],{"title":308,"text":309},"Your own data as part of the design","We start with the sources and users, not with an arbitrary model.\n",{"title":311,"text":312},"Integrations are part of it","The assistant can connect to existing software instead of becoming a new information silo.\n",{"title":314,"text":315},"Reliability over magic","Sources, uncertainty and limits may stay visible to the user.\n",{"intro":317,"signals":318},"The need is usually clear when a lot of valuable knowledge does exist, but is hard to find or scattered.\n",[319,322,325,328],{"title":320,"text":321},"Employees keep searching for the same information","Answers are spread across documents, systems and colleagues.\n",{"title":323,"text":324},"A general chatbot doesn't know your context","Good answers require access to organisation-specific information.\n",{"title":326,"text":327},"Sources change regularly","The solution has to work with current knowledge without baking everything into a model.\n",{"title":329,"text":330},"Not everyone may see everything","Roles, teams or clients have different access rights.\n",{"intro":332},"With Build & Integrate we build the assistant around your real sources, access model and usage situations. If it's still uncertain whether retrieval and answer quality will work well enough, we can first validate that core in an AI Prototype Sprint.\n",{"buyerTriggerKeys":334,"primaryOfferKey":93,"offerKeys":335,"caseKeys":336,"relatedUseCaseKeys":337},[117],[93,254],[],[121,338],"documenten-en-mails-automatisch-verwerken",{"title":340,"description":341},"Have an internal AI assistant developed | 9to5 software","Need an internal AI assistant or your own ChatGPT built on your knowledge and systems? 9to5 builds secure, integrated AI assistants for serious internal use.\n","use-cases\u002Finterne-ai-assistent-laten-ontwikkelen","You want to give employees faster access to internal knowledge, documents and systems through an assistant that fits your organisation.\n",{"hero":345,"problem":346,"proof":347},"sq-knowledge-filter","sq-connect-nodes","ls-system-connect","_peqkPNQXiEZoQx11qqN8kY_wuZT1mvO0j1tCcY3dno",{"id":350,"title":351,"body":352,"description":26,"desiredOutcome":356,"extension":53,"hero":368,"key":121,"meta":370,"navigation":69,"path":371,"problem":372,"proof":384,"recognition":396,"recommendedApproach":411,"relations":413,"seo":418,"stem":421,"summary":422,"visuals":423,"__hash__":424},"useCases_en\u002Fuse-cases\u002Feigen-data-en-kennis-gebruiken-met-ai.md","Make your own documents, data and knowledge usable for AI",{"type":23,"value":353,"toc":354},[],{"title":26,"searchDepth":27,"depth":27,"links":355},[],{"intro":357,"items":358},"You want to give AI access to relevant information of your own without putting everything into one uncontrollable data bin.\n",[359,362,365],{"title":360,"text":361},"Relevant information at the right moment","The AI only gets the context needed for the current task.\n",{"title":363,"text":364},"Sources stay traceable","Important output can be traced back to the underlying information.\n",{"title":366,"text":367},"Permissions and freshness stay intact","Users only see what they are allowed to see, and changes in sources can carry through into the solution.\n",{"intro":369},"A language model doesn't know your organisation by itself. If AI has to work with your own documents, product data or internal knowledge, that information has to be made available at the right moment and with the right permissions.\n",{},"\u002Fuse-cases\u002Feigen-data-en-kennis-gebruiken-met-ai",{"intro":373,"causes":374},"\"Giving our own data to AI\" sounds like one technical step, but usually consists of selection, retrieval, permissions and quality control.\n",[375,378,381],{"title":376,"text":377},"More context is not automatically better","A model actually gives worse answers when too much irrelevant information is included.\n",{"title":379,"text":380},"Sources have different structures","A PDF, a relational database and a ticketing system each call for a different approach.\n",{"title":382,"text":383},"Data permissions don't disappear with AI","Retrieval has to respect the same access that applies elsewhere in the organisation.\n",{"intro":385,"points":386},"The quality of an AI solution is often determined more by the context you provide than by the newest model. That's why we treat retrieval and data integration as fully-fledged engineering.\n",[387,390,393],{"title":388,"text":389},"Source and use case belong together","We don't just index everything because we can.\n",{"title":391,"text":392},"Existing permissions stay in charge","AI must not become a way around the access model.\n",{"title":394,"text":395},"No unnecessary platform layer","We only choose technical components the real problem requires.\n",{"intro":397,"signals":398},"This question arises when general AI works well, but business-specific context is missing.\n",[399,402,405,408],{"title":400,"text":401},"Knowledge is spread across many sources","Documents, databases and other systems each hold part of the context.\n",{"title":403,"text":404},"Copying into a prompt doesn't scale","Adding information by hand works for a demo, but not for structural use.\n",{"title":406,"text":407},"Information changes regularly","A static dataset quickly becomes outdated.\n",{"title":409,"text":410},"Not all knowledge is for everyone","Access rules from existing systems have to be preserved.\n",{"intro":412},"With Build & Integrate we build the technical layer that gives AI targeted access to your data and knowledge. If feasibility or answer quality is still uncertain, an AI Prototype Sprint can first validate the most important sources and use case.\n",{"buyerTriggerKeys":414,"primaryOfferKey":93,"offerKeys":415,"caseKeys":416,"relatedUseCaseKeys":417},[117],[93,254],[],[120,122],{"title":419,"description":420},"Use your own data and knowledge with AI | 9to5 software","Put your own documents and databases to work with AI? 9to5 builds RAG, retrieval, integrations and AI workflows around your existing information.\n","use-cases\u002Feigen-data-en-kennis-gebruiken-met-ai","The valuable information is already there. Now AI has to be able to work with it reliably, securely and practically.\n",{"hero":345,"problem":346,"proof":347},"rQE5CNfHzt2_sDfHXKyafJGMwiRmtAudJssA1bEegxQ",{"id":426,"title":427,"body":428,"description":26,"desiredOutcome":432,"extension":53,"hero":444,"key":122,"meta":446,"navigation":69,"path":447,"problem":448,"proof":460,"recognition":472,"recommendedApproach":487,"relations":489,"seo":494,"stem":497,"summary":498,"visuals":499,"__hash__":501},"useCases_en\u002Fuse-cases\u002Fai-toevoegen-aan-bestaande-software.md","Add AI to existing software, an app or SaaS",{"type":23,"value":429,"toc":430},[],{"title":26,"searchDepth":27,"depth":27,"links":431},[],{"intro":433,"items":434},"The new AI functionality should feel like part of the product, not like a chatbot glued on later.\n",[435,438,441],{"title":436,"text":437},"AI fits the existing user flow","The feature supports what users are already trying to do.\n",{"title":439,"text":440},"The existing architecture stays manageable","AI is integrated without needlessly adding a second technical world.\n",{"title":442,"text":443},"Production risks are covered","Costs, latency, errors and uncertain model behaviour get a place in the design.\n",{"intro":445},"You don't need to build a completely new AI product. Often most of the value lies in one smart feature inside software users already know. We make sure that feature fits the existing product technically.\n",{},"\u002Fuse-cases\u002Fai-toevoegen-aan-bestaande-software",{"intro":449,"causes":450},"Integrating a standalone AI API is easy. Designing AI as a reliable part of existing product software is another story.\n",[451,454,457],{"title":452,"text":453},"Existing software already has rules","Data, permissions, business logic and user expectations are largely fixed.\n",{"title":455,"text":456},"AI introduces other failure modes","Latency, incorrect output and variable costs have to be handled in existing flows.\n",{"title":458,"text":459},"The new feature can end up too detached","A separate AI system quickly leads to duplicated data and logic.\n",{"intro":461,"points":462},"Understanding and extending existing software is a big part of our work. So AI doesn't need a greenfield to be built well.\n",[463,466,469],{"title":464,"text":465},"Understand the existing architecture first","We design the AI feature around what already works.\n",{"title":467,"text":468},"Integration over isolation","Data, auth and business logic are not duplicated unnecessarily.\n",{"title":470,"text":471},"Production from the design onwards","Reliability and operational consequences count before the feature goes live.\n",{"intro":473,"signals":474},"The problem is usually not how to call a model, but how the new functionality should relate to the existing product.\n",[475,478,481,484],{"title":476,"text":477},"The feature needs existing product data","AI has to work with users, documents or other information already in the product.\n",{"title":479,"text":480},"Permissions must simply keep working","AI must not suddenly expose information or actions that are protected elsewhere.\n",{"title":482,"text":483},"The existing UX should stay in charge","Users want a better workflow, not necessarily a chat interface.\n",{"title":485,"text":486},"You don't want to rebuild the current product","The AI layer has to connect to what already works well technically.\n",{"intro":488},"With Build & Integrate we add the AI functionality to the existing codebase and systems in a targeted way. When the new component is technically risky or still uncertain in substance, an AI Prototype Sprint can first prove the core.\n",{"buyerTriggerKeys":490,"primaryOfferKey":93,"offerKeys":491,"caseKeys":492,"relatedUseCaseKeys":493},[117],[93,254],[],[257,121],{"title":495,"description":496},"Add AI to existing software or SaaS | 9to5 software","Add AI features (ChatGPT, Claude, OpenAI API) to existing software? 9to5 designs, builds and integrates AI into existing apps and SaaS products.\n","use-cases\u002Fai-toevoegen-aan-bestaande-software","The product already exists. Now you want to add AI functionality without losing sight of the existing architecture, user experience and reliability.\n",{"hero":346,"problem":500,"proof":347},"sq-foundation-build","uIjq2lxyg3PeXyGhq82YeTX4j_ALcHznMSqDHe95r9k",{"id":503,"title":504,"body":505,"description":26,"desiredOutcome":509,"extension":53,"hero":521,"key":123,"meta":523,"navigation":69,"path":524,"problem":525,"proof":537,"recognition":549,"recommendedApproach":564,"relations":566,"seo":576,"stem":579,"summary":580,"visuals":581,"__hash__":582},"useCases_en\u002Fuse-cases\u002Ftests-evals-monitoring-voor-ai.md","We're missing tests, evals or monitoring",{"type":23,"value":506,"toc":507},[],{"title":26,"searchDepth":27,"depth":27,"links":508},[],{"intro":510,"items":511},"You want to change things with confidence because you quickly see what a new prompt, model version or code change actually does.\n",[512,515,518],{"title":513,"text":514},"Regressions become visible","Important existing situations are automatically checked again.\n",{"title":516,"text":517},"AI quality becomes measurable","You compare changes against relevant examples and criteria.\n",{"title":519,"text":520},"Production gives feedback","Errors, latency and unusual behaviour no longer disappear quietly into logs.\n",{"intro":522},"\"It feels better\" is not a reliable test strategy. If you use AI functionality seriously, you need to be able to see what changes do and when things go wrong in production. We help set up that feedback loop.\n",{},"\u002Fuse-cases\u002Ftests-evals-monitoring-voor-ai",{"intro":526,"causes":527},"Traditional software tests remain important, but AI adds behaviour for which there isn't always one exactly correct answer.\n",[528,531,534],{"title":529,"text":530},"AI output is probabilistic","Exact string comparisons are not enough for many AI tasks.\n",{"title":532,"text":533},"Good examples aren't recorded","Knowledge of what is right and wrong mainly lives in heads and scattered chats.\n",{"title":535,"text":536},"Observability stops at technical errors","An HTTP 200 says nothing about the quality of the answer.\n",{"intro":538,"points":539},"A measurement layer is only useful to us if it helps engineers see faster whether a change is safe and actually better.\n",[540,543,546],{"title":541,"text":542},"Regular tests where possible","Deterministic software simply stays deterministically testable.\n",{"title":544,"text":545},"Evals where needed","AI behaviour gets examples and criteria that fit the real use case.\n",{"title":547,"text":548},"Production data closes the loop","Monitoring shows which real-world situations need attention.\n",{"intro":550,"signals":551},"Without a good feedback loop, an AI product becomes harder and harder to develop responsibly.\n",[552,555,558,561],{"title":553,"text":554},"Testing is mostly manual","After every change someone tries a few known examples.\n",{"title":556,"text":557},"Prompt changes are hard to compare","You can't objectively tell whether version B is really better than version A.\n",{"title":559,"text":560},"Production errors are discovered by users","The system itself hardly flags deviating behaviour.\n",{"title":562,"text":563},"Model updates feel risky","A new provider or model version can change behaviour without clear warning.\n",{"intro":565},"If the scope is clear, we set up a practical feedback loop in a Rescue Sprint with the right combination of tests, evals, traces and monitoring. If broader research is needed first, we start with an audit.\n",{"buyerTriggerKeys":567,"primaryOfferKey":569,"offerKeys":570,"caseKeys":572,"relatedUseCaseKeys":573},[568],"vastgelopen-met-ai","rescue-sprint",[569,571],"vibe-code-rescue-audit",[],[574,575],"ai-agent-niet-betrouwbaar","ai-code-breekt-steeds",{"title":577,"description":578},"Tests, evals and monitoring for AI | 9to5 software","AI without proper tests, LLM evals or monitoring? 9to5 helps set up measurable quality, regression tests and production observability.\n","use-cases\u002Ftests-evals-monitoring-voor-ai","The AI functionality runs, but nobody can objectively see whether changes make it better, worse or quietly break something.\n",{"hero":266,"problem":265,"proof":181},"HLgbDqrGnq-G64ciWMNMDChkwkuEQEyI9a7eCclpOe0",1792049913642]