The Methodology
A framework for installing a new operating model in two weeks. Built from twenty-five years of senior product practice, encoded into an agent system that travels with your team after the residency ends.
The premise
the missing layer
Companies are buying AI tools and watching adoption metrics. The tools work. The metrics rise. Output does not. The problem is structural, not technological. The operating model that organizes how decisions get made, how trade-offs surface, and how judgment gets exercised was designed for a workforce that no longer applies.
The companies that are actually shipping at the new rate share a pattern. A small team of senior practitioners, paired with an agent layer that absorbs the coordination tax, operating under a shared paradigm that survives without enforcement. This is the operating model AI-native work assumes. Most companies are running an older model and wondering why their output curve is flat.
Marliin's methodology installs the new operating model. The two-week residency teaches your team how to inhabit it. The agent system encodes it so the model travels home and runs continuously in your environment. The teaching is what your team needs. The agents are what the methodology needs to compound.
The three forces that compress when AI enters the operating model.
Pattern recognition that used to take a decade now takes a project.
A junior practitioner used to need years of real exposure to develop instinct. Patterns came from seeing the same situation play out dozens of times, slowly, across engagements. With agents running scenarios on real company data, a practitioner can see a hundred versions of a decision in a week, with structured feedback on each one. The bottleneck on judgment was never difficulty. It was bandwidth. AI removes the bandwidth ceiling.
Tacit judgment becomes legible by the act of directing the system.
When a senior practitioner directs an agent, they have to externalize judgment that would otherwise have stayed tacit. The instincts that took decades to absorb by osmosis become prompts, overrides, rejected suggestions, written-down reasons. Mentorship was always capped by the senior's willingness and ability to explain their thinking. Now the explanation is a forced byproduct of getting any work done. The team learns from artifacts that seniors could not previously produce.
Real work becomes safe to ship under continuous review.
Junior work used to be either real and dangerous, or fake and worthless. With an agent system acting as continuous first reviewer and a senior practitioner as occasional calibrator, a practitioner can ship real work with a safety layer that did not exist before. Failure is fast, cheap, and observable. The trust gap that used to take years to close now closes in weeks.
Fourteen days. Four movements. One operating model installed.
Orient
Days 1–2
Frame the real problem your team brought. Establish shared vocabulary. Set operating principles.
Pair
Days 3–7
Daily paired work with senior Marliin practitioners. The agent system learns by watching your team decide.
Compose
Days 8–11
Your team begins directing the agents under observation. Coaching shifts from active to observational.
Carry
Days 12–14
Hand-off. The agent system migrates to your environment. Post-residency cadence designed so the model holds at home.
The asset
what compounds
The methodology is not abstract. It is encoded into an agent system that we have been building since Marliin's first engagement. We call this system the Agent Core. Every client residency adds to it. Every senior decision overridden in real work feeds it. Every framework refined during a hand-off becomes part of it.
When your team arrives at the residency, the Agent Core is what the senior practitioners are pairing you with. When you leave, a custom instance of the Agent Core, trained on your domain during the residency, travels home with your team. It runs in your environment continuously. It holds the operating model against your organization's antibodies.
The teaching is what your team needs. The Agent Core is what the methodology needs to compound. The hundredth client benefits from everything the previous ninety-nine taught the system.
The Agent Core is built on Marliin's accumulated practice. It is the durable asset that residency clients license. It is also what makes the methodology venture-shaped: a software layer that gets more valuable with each engagement, not a service that resets to zero every time.
Pivotal Labs proved that small teams of high-judgment people ship dramatically better. The cap on that model was the cost of getting humans aligned. The cap just came off.
Brett Marlin · Founder, Marliin
Currently embedded with a defense AI company building autonomous fleet sustainment.
The first commercial proof of the methodology. The Agent Core trained for analyst decision-making in classified environments.
Two paid pilots planned for the second half of 2026 in leased Northern California properties.
Three to four teams per pilot. The pilots de-risk the residency model before the permanent property is committed.
Cohort I opens in Northern California, 2027.
Limited cohort, selected for fit. Applications and waitlist open now.
In context
Marliin is not alone in seeing this shape. The market is now writing about what we are teaching.
In May 2026, the Wall Street Journal reported that Coinbase, AWS, and Ernst & Young are all describing the same emerging structure: small, cross-functional teams of one to eight humans operating with AI agents, replacing engineering groups of ten to fifteen. The new word for these teams is pods. Coinbase's head of platform, Rob Witoff, attributed the gains to hyper-aligned people with very little unnecessary overhead.
The same month, a controlled study from Carleton University and Defence R&D Canada measured the architectural question directly. Across 3,475 episodes and six models, the largest performance gains came not from deeper reasoning but from what the paper calls context engineering. Programmatic state abstraction that compresses what each agent sees before deliberation begins. The improvement: up to 76% better outcomes per token spent. The paper also documented a destructive pattern it named the deliberation cascade. Distributing more reasoning across more agents degraded performance by up to 3.4× while consuming 1.8 to 2.7× more tokens. The conclusion was unambiguous: invest in programmatic infrastructure and clean task decomposition rather than deeper per-agent reasoning.
That is the methodology Marliin teaches, restated as an empirical finding. The Agent Core is the deterministic infrastructure the paper recommends. Context Engineering, named identically in our methodology and theirs, is the practice the paper measures as the highest-leverage architectural choice. Workflow Cartography produces the bounded task decomposition the paper validates. Role Reframing produces the bounded specialists. The deliberation cascade the paper names is the failure mode of AI mandates that bolt more reasoning onto more agents without engineering the context they operate in. Companies trying to figure this out by trial and error will produce the failure mode the paper documented. The Residency installs the architecture that avoids it.
The pod is the destination. The Residency is how teams arrive there without spending two years discovering by trial and error what the field has now measured directly.
"Your Work Team Is Now a 'Pod' and Your Co-Workers Are AI Agents"
Wall Street Journal CIO Journal · May 2026
Bogdanov, Lung, Kunz, Gao, Taylor, Zaman · ACM CAIS · May 2026
For your team
If your team is serious about installing a new operating model, not just buying tools, start the application process.
Apply →For investors
Marliin Residency is venture-backable on the basis of the Agent Core, not the building. Investor inquiries welcome.
Investor inquiries →