[{"data":1,"prerenderedAt":183},["ShallowReactive",2],{"entry-\u002Fblog\u002Fdoes-your-outsourcing-company-really-use-ai":3,"entry-related-\u002Fblog\u002Fdoes-your-outsourcing-company-really-use-ai":153},{"path":4,"title":5,"description":6,"cover":7,"published":10,"updated":11,"tags":12,"translated":15,"seo":16,"body":17},"\u002Fblog\u002Fdoes-your-outsourcing-company-really-use-ai","Does Your Outsourcing Vendor Really Use AI?","AI-washing is easy to fake on a slide. Here's the due-diligence checklist that separates a vendor's real tooling from a deck with a robot icon on it.",{"src":8,"alt":9},"\u002Fmedia\u002F2026\u002F09\u002F8da6b526d694a704.webp","Four people gathered around a laptop on a glass table during a discussion","2026-09-24",null,[13,14],"AI Trends","Software Development",false,{"title":5,"description":6},{"type":18,"value":19,"toc":145},"minimark",[20,34,57,62,77,80,84,87,96,100,103,106,110,113,121,130],[21,22,23,27,28,33],"p",{},[24,25,26],"strong",{},"The only way to tell a vendor that has rebuilt its process around AI from one that added a slide about it is to ask questions a slide can't answer."," Every outsourcing pitch in 2026 claims ",[29,30,32],"a",{"href":31},"\u002Fblog\u002Fhow-ai-is-reshaping-software-outsourcing-in-2026\u002F","an AI-driven process",", because the phrase costs nothing to say and nothing to put in a proposal. The claim stopped being a differentiator the moment it stopped being checkable, so the burden has shifted to the buyer: know what to ask for, and know what a real answer sounds like versus a rehearsed one.",[35,36,38],"key-points",{"label":37},"Key points",[39,40,41,45,48,51,54],"ul",{},[42,43,44],"li",{},"Ask which parts of the delivery pipeline are actually agent-driven and which are human-reviewed, and get a specific point in the workflow where review happens — not a general assurance that \"a person checks everything.\"",[42,46,47],{},"Ask for one metric that moved because of the tooling, with a real baseline from before it was adopted. A percentage with no starting point is a marketing number, not an engineering one.",[42,49,50],{},"Ask to see the tooling running against real work, not a slide describing it. A vendor with nothing to demo has nothing running.",[42,52,53],{},"Ask what happens the last time the AI tool was wrong. The answer reveals whether a human is actually in the loop or the process just trusts the output.",[42,55,56],{},"A vendor with real signal answers all four specifically and quickly; a vendor that's AI-washing reaches for the same three buzzwords regardless of which question you asked.",[58,59,61],"h2",{"id":60},"where-the-pipeline-is-agent-driven-and-where-a-human-takes-over","Where the pipeline is agent-driven, and where a human takes over",[21,63,64,65,69,70,76],{},"The most useful single question in a vetting call is also the simplest: which of your steps does the agent do unsupervised, and which does a person redo or check? A vendor with a real process can draw this line immediately, because they had to design it — deciding where an agent's output ships without a second look and where it doesn't is an actual engineering decision with consequences either way — the same ",[29,66,68],{"href":67},"\u002Fblog\u002Fai-can-write-code-why-companies-still-outsource\u002F","judgment gap that AI never eliminated",", just moved to a different point in the pipeline. They'll name the boundary in specifics: agents scaffold CRUD endpoints and generate test cases without review, but a senior engineer ",[29,71,75],{"href":72,"rel":73},"https:\u002F\u002Fen.wikipedia.org\u002Fwiki\u002FCode_review",[74],"nofollow","reviews every diff"," that touches auth, payments, or a database migration before it merges.",[21,78,79],{},"A vendor that's AI-washing gives you the opposite answer, usually in the opposite order of specificity. They'll say something like \"our engineers work closely with AI tools throughout the process,\" which is true of almost every dev shop on earth and describes nothing. Push once — \"which step, specifically, does the agent do without a human re-checking it\" — and listen for whether the answer gets more concrete or more vague. A team that has actually built the boundary gets more concrete under a follow-up question. A team that hasn't retreats further into the general.",[58,81,83],{"id":82},"a-metric-that-moved-measured-from-a-baseline","A metric that moved, measured from a baseline",[21,85,86],{},"\"AI made us faster\" is not a claim you can evaluate. \"Our cycle time from ticket to merged PR dropped from four days to a day and a half after we moved code review augmentation into the pipeline in Q1\" is. The difference isn't the size of the number — it's that the second claim names what was measured, when the baseline was taken, and what specifically changed between the before and after. That's the shape of a real answer, whether the number is dramatic or modest.",[21,88,89,90,95],{},"Ask for the metric a vendor actually tracks internally, not the one that sounds best in a sales conversation. Cycle time, defect escape rate, and time-to-first-demo are the three worth asking about directly, because all three are things a delivery team would be measuring anyway if they were serious about their ",[29,91,94],{"href":92,"rel":93},"https:\u002F\u002Fen.wikipedia.org\u002Fwiki\u002FDevOps",[74],"DevOps"," practice — AI or not. If a vendor can produce one of these with a before-and-after and a rough date, that's a team that instruments its own work. If every answer is a percentage with no anchor — \"40% faster,\" \"60% fewer bugs\" — with no baseline offered even when you ask for one, treat the number as a decoration rather than a data point.",[58,97,99],{"id":98},"ask-to-see-it-not-hear-about-it","Ask to see it, not hear about it",[21,101,102],{},"A slide can claim anything. A live screen share of the actual repository, the actual agent runs, and the actual review comments cannot. This is the highest-signal request in the entire vetting process, because it's the one a vendor can't fake with better writing: either they can pull up a real pull request an agent opened last week and show you what a reviewer flagged on it, or they can't.",[21,104,105],{},"What you're looking for in that walkthrough isn't polish — it's ordinariness. A real workflow has messy commits, agent output that got rejected and redone, and review comments that read like they were written by someone frustrated at 4pm, not by someone performing competence for an audience. If everything you're shown looks curated for the demo, ask to see something that wasn't prepared in advance — last week's actual PR history, not a hand-picked example. A vendor with nothing to hide has no reason to resist that.",[58,107,109],{"id":108},"what-happens-when-the-tool-is-wrong","What happens when the tool is wrong",[21,111,112],{},"Every AI tool used in production is sometimes wrong, and how a vendor talks about that fact tells you more than almost anything else on this list. A team with a real process has a story ready, because they've lived it: an agent confidently proposed a database index that would have degraded a hot query path, and a senior engineer caught it in review before it shipped. That's not a failure story — it's evidence the review step is real rather than theoretical.",[21,114,115,116,120],{},"A vendor that hasn't thought about this will either claim their tooling doesn't make mistakes, which should end the conversation on its own, or give you a vague answer about \"human oversight\" with no example attached. If nobody on the other side of the table can describe a specific time the AI was wrong and what stopped it from shipping, there's a good chance nobody has actually been checking — and that gap is also a contract question: ",[29,117,119],{"href":118},"\u002Fblog\u002Foutsourcing-vendor-ai-code-who-owns-the-risk\u002F","who actually owns the risk"," when a defect like that ships anyway.",[21,122,123,124,129],{},"None of these four questions go beyond ordinary ",[29,125,128],{"href":126,"rel":127},"https:\u002F\u002Fen.wikipedia.org\u002Fwiki\u002FDue_diligence",[74],"due diligence"," — they don't require the vendor to hand over proprietary information or break a client NDA — a specific workflow boundary, one metric with a baseline, a real screen share, and one failure story are all things a team with a genuine process can produce without exposing anything sensitive. That's part of what makes them useful filters: the vendors who dodge aren't protecting secrets, they're protecting the absence of an answer.",[21,131,132,133,138,139,144],{},"If you want to run this checklist against how we actually work, ",[29,134,137],{"href":135,"rel":136},"https:\u002F\u002Fbigin.vn\u002Fcontact-us\u002F",[74],"get in touch"," — or read more about how we've built our delivery process on ",[29,140,143],{"href":141,"rel":142},"https:\u002F\u002Fbigin.vn\u002Fabout-us",[74],"our about page",".",{"title":146,"searchDepth":147,"depth":147,"links":148},"",2,[149,150,151,152],{"id":60,"depth":147,"text":61},{"id":82,"depth":147,"text":83},{"id":98,"depth":147,"text":99},{"id":108,"depth":147,"text":109},{"tag":13,"entries":154},[155,164,173],{"path":156,"title":157,"description":158,"cover":159,"published":162,"tags":163},"\u002Fblog\u002Foutsourcing-red-flags-companies-discover-too-late","The Outsourcing Red Flags Most Companies Discover Too Late","The vendor red flags that sink projects are checkable before you sign — most buyers only look for them after the damage is done.",{"src":160,"alt":161},"\u002Fmedia\u002F2026\u002F09\u002F0e798592791e5298.webp","A red warning flag on a pole by the shoreline","2026-10-04",[14],{"path":165,"title":166,"description":167,"cover":168,"published":171,"tags":172},"\u002Fblog\u002Fsmall-ai-augmented-team-vs-20-person-team","Small AI-Augmented Team vs. 20-Person Dev Team","A lean AI-augmented team wins on speed and loses on breadth. Here's how to size the trade-off before you commit to either structure.",{"src":169,"alt":170},"\u002Fmedia\u002F2026\u002F09\u002F44e99321e960f1dc.webp","Three colleagues sitting together at a table in an office","2026-10-02",[13,14],{"path":174,"title":175,"description":176,"cover":177,"published":180,"tags":181},"\u002Fblog\u002Ffirst-30-days-with-an-outsourcing-partner","The First 30 Days with an Outsourcing Partner","What a well-run first month of outsourced delivery looks like week by week, and the early signals that tell you whether the engagement is on track.",{"src":178,"alt":179},"\u002Fmedia\u002F2026\u002F09\u002F47838004b21ebdfa.webp","Three people working on laptops while a colleague presents at a whiteboard","2026-09-30",[14,182],"Product Development",1792485816207]