[{"data":1,"prerenderedAt":296},["ShallowReactive",2],{"insights-index":3},[4,176],{"id":5,"title":6,"author":7,"authorRole":8,"body":9,"date":157,"description":158,"extension":159,"featured":160,"heroImage":161,"image":161,"imageAlt":161,"kind":162,"meta":163,"navigation":164,"path":165,"practice":166,"readingTime":167,"seo":168,"stem":169,"tags":170,"__hash__":175},"insights\u002Finsights\u002Fmaking-agents-accountable.md","Accountability Is the Hard Part of Agentic Engineering","Lunari","Founding Team",{"type":10,"value":11,"toc":143},"minimark",[12,16,21,24,27,33,37,42,45,53,57,60,63,67,74,77,81,84,88,91,98,102,105,121,124,127],[13,14,15],"p",{},"Every enterprise AI pilot demos well. A striking number never reach production. The\ngap between the two is the absence of the machinery that makes an autonomous\nactor safe to trust with consequential work.",[17,18,20],"h2",{"id":19},"the-demo-to-production-gap","The Demo-to-Production Gap",[13,22,23],{},"A demo is a single happy path executed once, under supervision, on curated data. Production\nis thousands of executions across months, on data nobody cleaned, with staff turnover,\nchanging permissions, and an auditor who will eventually ask what happened on a specific\nTuesday in March.",[13,25,26],{},"Those are different engineering problems. The first needs a capable model. The second needs\ninfrastructure that most organizations have not built, because until recently there was no\nreason to.",[28,29,30],"blockquote",{},[13,31,32],{},"An agent that cannot be audited cannot be trusted with work that matters. Accountability\nis the substrate everything else stands on, and it has to be there from the start.",[17,34,36],{"id":35},"four-things-that-have-to-exist","Four Things That Have to Exist",[38,39,41],"h3",{"id":40},"memory-that-persists","Memory That Persists",[13,43,44],{},"Context windows are working memory, not institutional memory. When a session ends, an agent\nwithout persistent storage forgets everything — the decision it made, the reason it made it,\nand the correction you gave it. Every session restarts from zero.",[13,46,47,48,52],{},"Real memory means facts that survive across sessions, models, and vendors, with the\nprovenance of each belief attached. It also means being able to reconstruct what the agent\nbelieved at a past moment, separately from what it believes now — because when something\ngoes wrong, the question is always ",[49,50,51],"em",{},"what did it know at the time",".",[38,54,56],{"id":55},"identity-that-travels","Identity That Travels",[13,58,59],{},"If an agent's identity lives inside one vendor's product, you do not have an agent — you\nhave a subscription. The role, its accumulated knowledge, and its track record should be\nportable across the models that execute it.",[13,61,62],{},"This matters commercially as much as technically. Model capability is converging and prices\nare falling. Betting your operating model on a specific provider's proprietary agent\nabstraction is a strategic risk as much as an architectural one.",[38,64,66],{"id":65},"permission-that-actually-holds","Permission That Actually Holds",[13,68,69,70,73],{},"Most agentic frameworks treat permissions as a prompt instruction: ",[49,71,72],{},"do not modify records\nwithout approval",". That is a request, made to a non-deterministic system, that will\neventually be declined.",[13,75,76],{},"Real permissioning is enforced outside the model, at the point of execution. The agent\nproposes; a deterministic layer decides whether the action is allowed, given who the agent\nis acting for and what it has been authorized to do. The model cannot talk its way past a\nsystem it does not control.",[38,78,80],{"id":79},"audit-that-stands-up","Audit That Stands Up",[13,82,83],{},"When an agent takes an action, six months later someone will need to know: what did it do,\non whose authority, based on what belief, and could it have been prevented. If the answer\nlives in an ephemeral chat log, the organization cannot defend the decision.",[17,85,87],{"id":86},"earned-autonomy","Earned Autonomy",[13,89,90],{},"The pattern we keep returning to is incremental trust. An agent begins with narrow scope and\nsupervised execution. As it demonstrates reliability on a class of work, its scope widens.\nFailures narrow it again.",[13,92,93,94,97],{},"This mirrors how organizations onboard people, and for the same reason: capability has to\nbe demonstrated before it is relied on. It also gives leadership a defensible answer to the question\nthat stalls most deployments — ",[49,95,96],{},"how do you know it will not do something catastrophic?"," The\nanswer is that it has not been granted the ability to, and will not be until it has earned it.",[17,99,101],{"id":100},"what-this-means-practically","What This Means Practically",[13,103,104],{},"If you are evaluating agentic systems, the questions worth asking are less about the model\nand more about the substrate:",[106,107,108,112,115,118],"ul",{},[109,110,111],"li",{},"Where does the agent's memory live, and who can audit it?",[109,113,114],{},"If you changed model providers next quarter, what would you lose?",[109,116,117],{},"Are permissions enforced in code, or requested in a prompt?",[109,119,120],{},"Can you reconstruct why the agent acted as it did, months after the fact?",[13,122,123],{},"A vendor with strong answers to those has built infrastructure. A vendor lacking them has\nbuilt a demo, and closing that gap will consume the budget you allocated for results.",[125,126],"hr",{},[13,128,129,132],{},[49,130,131],{},"Lunari builds agentic enterprise platforms and advises leadership teams deploying them.\nIf this maps to a problem you are working on, ",[49,133,134,142],{},[135,136,137],"strong",{},[138,139,141],"a",{"href":140},"\u002Fcontact","we should talk","**.",{"title":144,"searchDepth":145,"depth":145,"links":146},"",3,[147,149,155,156],{"id":19,"depth":148,"text":20},2,{"id":35,"depth":148,"text":36,"children":150},[151,152,153,154],{"id":40,"depth":145,"text":41},{"id":55,"depth":145,"text":56},{"id":65,"depth":145,"text":66},{"id":79,"depth":145,"text":80},{"id":86,"depth":148,"text":87},{"id":100,"depth":148,"text":101},"August 9, 2026","Model capability stopped being the bottleneck. What blocks agents from real enterprise work is everything around the model — memory, identity, permission, and audit. Here is how we think about building it.","md",false,null,"article",{},true,"\u002Finsights\u002Fmaking-agents-accountable","advisory",8,{"title":6,"description":158},"insights\u002Fmaking-agents-accountable",[171,172,173,174],"agentic engineering","governance","enterprise AI","subcortex","fazrZ7kdzt8tyICgZpa09iiqIgwaZr9LzRvtpmAs6dk",{"id":177,"title":178,"author":179,"authorRole":180,"body":181,"date":284,"description":285,"extension":159,"featured":164,"heroImage":161,"image":286,"imageAlt":287,"kind":162,"meta":288,"navigation":164,"path":289,"practice":166,"readingTime":290,"seo":291,"stem":292,"tags":293,"__hash__":295},"insights\u002Finsights\u002Fprocess\u002Fthe-question-that-decides-whether-your-agentic-strategy-fails.md","The Question That Decides Whether Your Agentic Strategy Fails","Tom Santos","President and CEO",{"type":10,"value":182,"toc":280},[183,186,189,192,196,199,243,246,249,252,255,259,262,265,268,271,273],[13,184,185],{},"If you're getting ready to build with agents or if you're struggling to see the results you keep hearing about, let me ask you the one question that will determine whether your agentic engineering strategy works:",[13,187,188],{},"Are you starting with the process, or with the model?",[13,190,191],{},"I've watched both paths play out. I spent more than a decade consulting on exactly this, inside startups, mid-market companies, and global enterprises, brought in to fix processes that were producing bad outcomes. The pattern held across every engagement: the fix lived in one of a handful of core phases, and it had nothing to do with the platform they'd paid for or the framework they'd just rolled out. Today my teams ship applications and features through an AI-PDLC every day, and I can tell you the choice you make in your first week will shape everything that follows.",[17,193,195],{"id":194},"the-process-pieces-have-never-changed","The Process Pieces Have Never Changed",[13,197,198],{},"Here's what's true about application development: Frameworks have cycled for decades. Waterfall gave way to Agile, Agile spawned a dozen flavors, while the durable phases underneath stayed put:",[200,201,202,208,214,220,226,232,237],"ol",{},[109,203,204,207],{},[135,205,206],{},"ENVISION"," the need.",[109,209,210,213],{},[135,211,212],{},"GATHER"," the requirements that solve the correct problem.",[109,215,216,219],{},[135,217,218],{},"DESIGN"," the concept and design the architecture.",[109,221,222,225],{},[135,223,224],{},"BUILD"," the product.",[109,227,228,231],{},[135,229,230],{},"TEST"," and verify the product.",[109,233,234,225],{},[135,235,236],{},"RELEASE",[109,238,239,242],{},[135,240,241],{},"SUPPORT"," the users.",[13,244,245],{},"Frameworks decide how you slice those phases and how fast you loop them. The phases themselves are the work, and they form a chain. Your outcome is governed by the weakest link in it.",[13,247,248],{},"Some teams did improve when they moved to Agile, because long feedback loops were the wrong fit for products with uncertain requirements. The framework change fixed a fit problem, and that's real. What it left untouched was practice. Teams that ran thin requirements and rushed design under Waterfall carried those habits into their sprints, and the work broke down for the same reasons it broke down before.",[13,250,251],{},"Many of those teams are looking at agentic development now and wondering why they're still struggling, and the truth is agents amplify whatever process you hand them — good and bad. Give them a strong specification and they produce work you can ship. Give them a vague ticket and they produce volume you have to untangle.",[13,253,254],{},"And agents change the economics of the chain. Under human teams, build carried most of the cost, which is why requirements and testing were the first things cut under a deadline. When build collapses toward zero, the cost moves to specification and verification, precisely the links weak teams were already skipping.",[17,256,258],{"id":257},"start-with-the-process","Start With the Process",[13,260,261],{},"So the work starts before an agent is ever engaged. Take the chain and audit it honestly, phase by phase. Strengthen the chain, and you strengthen your agentic output.",[13,263,264],{},"Tokens accelerate a process, but they also inherit its gaps. A team that struggles to elicit real user needs will generate elegant solutions to imagined problems. A team that skips technical design will accumulate architecture nobody can defend and potentially a nightmare to support. A team without a verification strategy will ship faster and learn about the consequences from its users.",[13,266,267],{},"The tools and frameworks will keep changing, but the key phases of any application development process haven't changed in decades. Your results will keep tracking the weakest links in your chain. So find them.",[13,269,270],{},"Start with the process.",[125,272],{},[13,274,275],{},[49,276,277,278,52],{},"Lunari builds agentic enterprise platforms and advises leadership teams deploying them. If this maps to a problem you are working on, ",[138,279,141],{"href":140},{"title":144,"searchDepth":145,"depth":145,"links":281},[282,283],{"id":194,"depth":148,"text":195},{"id":257,"depth":148,"text":258},"2026-08-17","Agents amplify whatever process you hand them. After a decade spent fixing broken delivery processes, one question predicts whether an agentic strategy will produce results — are you starting with the process, or with the model?","\u002Fimages\u002Finsights\u002Fprocess-before-model.jpg","The seven durable phases of software delivery — envision, gather requirements, design, build, test, release, support — drawn as a connected chain of gears",{},"\u002Finsights\u002Fprocess\u002Fthe-question-that-decides-whether-your-agentic-strategy-fails",5,{"title":178,"description":285},"insights\u002Fprocess\u002Fthe-question-that-decides-whether-your-agentic-strategy-fails",[294],"process","NQ1qgUmizlV_WvHA568NpuzrAM2O-co-59_jH46PY40",1787025335751]