Executive summary
Q2 was the quarter where the agent work stopped being a collection of experiments and became part of Lifely’s operating layer.
Stephane worked across the places where Lifely had the most friction: preorder visibility, customer service, hiring, marketing execution, reporting trust, expansion research, and the internal scripts that connect platforms together.
The value was not one deliverable. It was a set of connected systems, coaching loops, and technical foundations designed to make Lifely less dependent on ad-hoc fixes and more capable of operating with durable internal infrastructure.
The shape of the quarter
The work covered more than one department. The useful way to read Q2 is by durable systems created, operational layers strengthened, and capabilities transferred to the team.
Workstreams
These are the main touchpoints a Lifely reader should recognize without needing internal context or technical shorthand.
PO Allocation and Before Ship
Stephane worked on the logic that connects purchase orders, preorder promises, customer expectations, and before-ship decisions. This meant clarifying what the team needed to know before communicating with customers, before changing orders, and before deciding what stock should be allocated where.
Hiring OS
Hiring became a structured operating system rather than a loose flow of applications and candidate opinions. The work covered application intake, candidate evaluation, scorecards, review workflows, tests, interviews, and the process the hiring team could use to move from raw applicants to clearer decisions.
Maurice
Maurice became the HR agent layer around the Hiring OS. The work involved shaping how the agent reads applications, summarizes candidates, supports evaluation, prepares screening, and gives the team a consistent way to interact with hiring information.
Cornelia
Cornelia represents the customer service agent track: helping Lifely turn customer questions, order context, and internal knowledge into a more structured support workflow. The goal was not only to answer tickets faster, but to give customer service a better operating brain.
Media buying dashboard and visibility
Stephane worked on dashboard trust, reporting definitions, revenue logic, ad spend visibility, and the link between media buying activity and business metrics. A major part of the work was making sure the numbers were understandable enough to be acted on.
Creative skills and execution systems
The media buying work included skill-building and tooling around static ads, ad hooks, UGC scripts, creative research, testing structure, and the handoff between strategy and execution. This made the creative process less dependent on scattered conversations and more dependent on reusable systems.
Little Lifely Europe expansion
Stephane worked on the Europe expansion track as its own project: country opportunities, market assumptions, page and offer logic, product sizing questions, and the systems needed to turn expansion research into operational decisions.
Fake-door product testing
Separate from the Europe expansion work, Stephane also supported fake-door testing for new product demand. This involved shaping the test logic, offer structure, internal scripts, and interpretation layer so Lifely could validate interest before committing to heavier product or inventory decisions.
Market, channel, and creative research
Stephane ran several research tracks that shaped Lifely’s marketing direction: TikTok Shop research using Comfort as a reference brand, Road to One Billion research, Yapper ads, future email marketing agents, and virality video research. These were not isolated notes, but inputs for how Lifely could build better marketing systems and creative workflows.
Impeccable and reusable design skills
Stephane implemented design skills such as Impeccable so Caesar and the broader agent system could produce cleaner, more consistent interfaces and reports instead of relying on generic output.
Caesar platform awareness
A major part of Q2 was setting up platform skills so Caesar could understand the tools available to him and how to operate across them. This included the broader architecture around Cloudflare, Cin7, Supabase, VPS-hosted services, and connected platform heads.
Voice agent, assessments, and skill tests
The HR tool included a substantial project around creating a voice agent and rebuilding the evaluation layer from scratch: personality assessments, candidate interpretation, and skill-test logic that could support hiring decisions without becoming a loose manual process.
BRO skill and PostHog-based optimization
Stephane created the BRO skill around Banana Rate Optimization, Lifely’s customer conversion rate optimization system. This included figuring out how to use PostHog, connect the data and workflows, and make the system usable by the team rather than keeping optimization knowledge trapped in one person’s head.
SEO skill for reporting and marketing work
Stephane also created an SEO skill to support Lifely’s reporting and marketing needs, giving the agent system a more structured way to approach search-oriented analysis and content direction.
Full dashboard accuracy review
Beyond individual dashboard fixes, Stephane reviewed dashboard accuracy more broadly: checking whether the numbers, definitions, sources, and business interpretations were reliable enough to support decisions.
Listicle system and ad deployment
For Little Lifely, Stephane created a reusable listicle template, deployed six listicles, and supported the deployment of roughly one hundred ads. This connected content structure, creative testing, and paid traffic execution into a repeatable system.
Caesar operations, maintenance, and scale
Caesar was not only a coordination layer. Q2 included the work of discovering how to operate an agent at scale: maintenance systems, deployment discipline, operating procedures, monitoring habits, and the permissions/security layer needed to let an agent work across the business without creating uncontrolled risk.
Backend, permissions, and internal scripting
Behind the visible agent work, Stephane worked on the technical foundations that make long-term systems possible: internal scripts, data flows, platform connections, deployment choices, security rules, and clear ownership boundaries.
Permissions and controlled access
A major Q2 thread was the authorization and permission layer around agent work: who can trigger what, which tools are available to which workflows, what needs approval, and how permissions are stored and enforced across Slack and Discord through a database-backed system.
Cloudflare, Supabase, VPS, and connected services
The architecture work covered Cloudflare-hosted pages and workers, Supabase-backed data layers, VPS-hosted services, Cin7 and Shopify connections, customer service systems, inventory tools, and order systems. This is the infrastructure that lets agents work with real operational context.
Making business-critical scripts safer
Several Q2 threads focused on scripts that connected important operational platforms. The work was about clearer inputs, clearer outputs, guarded write paths, logs, documentation, and ownership so the team could maintain them instead of relying on fragile black boxes.
Operating agents as infrastructure
Part of the quarter was learning what it takes to run agents like infrastructure: maintenance routines, deployment patterns, error recovery, monitoring, permission checks, and the discipline needed for agents to support daily operations safely.
Coaching and capability transfer
A large part of Q2 was not only building systems. It was coaching Lifely’s team on how to use them, reason about them, and eventually create similar systems themselves.
Using agents as systems, not chatbots
Stephane coached the team on how to brief agents, provide context, evaluate outputs, and turn repeated questions into reusable workflows. The emphasis was on building judgement, not just getting answers from an AI tool.
Turning business needs into system priorities
Regular strategy conversations helped clarify which systems mattered, what should be built first, where automation was safe, and where better infrastructure was needed before the team could rely on agents operationally.
Designing workflows people can actually run
The coaching work translated operational questions into clearer workflows: what context is needed, what the agent can safely do, where human judgement remains necessary, and how the team should maintain the system over time.
Applying AI to media buying and creative work
Stephane also coached the media buying and creative side on static ads, hooks, UGC scripts, testing angles, and how to use agent skills without turning the output generic.
Quarter arc
Mapping the operating problems
Q2 opened with strategy, access, reporting trust, hiring direction, expansion research, and the first serious work around the systems behind preorder and operational visibility.
Turning workflows into systems
PO Allocation, Before Ship, dashboard repair, Hiring OS, Maurice, and media buying enablement moved from discussion into structured workflows, tools, and clearer team habits.
Strengthening the foundations
The final month concentrated on making the work more operational: internal scripts, backend architecture, live system gates, HR workflows, voice screening, expansion testing, and clearer ownership across the agent stack.
Challenges
The work surfaced several constraints that matter for the next phase. These are not failures of the agent work. They are the operating realities that need to be solved if the systems are going to scale.
Understanding the data flow takes deep work
A large part of Q2 went into understanding how the operational backend actually works: where data starts, how it moves, which systems hold the truth, and where the gaps appear. This work does not always create immediate visible output, but it is necessary before reliable systems can be built on top of it.
No single senior engineering owner
Lifely is building with the technical visibility and resources available. Because there is not one dedicated senior engineer overseeing the whole architecture, Stephane often had to step into that role: mapping systems, asking technical questions, translating between business needs and implementation reality, and helping create structure where ownership was unclear.
Caesar is scaling without a playbook
Caesar has worked so far, but operating an agent across more users and more workflows is still discovery work. Every new use case creates new questions around permissions, context, maintenance, routing, error handling, and operational discipline. The current setup is functional, but continued growth will create new problems to solve.
Front-line marketing context is harder to track asynchronously
Because Stephane is not in every marketing meeting and the team operates across time zones, it is harder to keep a full view of what is happening on the front lines. A centralized update system, meeting summary layer, or internal newsletter could help keep business context visible without adding more live meetings.
Next quarter priorities
The highest-value work in Q3 is not only more marketing output. Marketing can create demand, but Lifely also needs the backend infrastructure and customer experience layer to support that demand safely.
Map and stabilize the backend infrastructure
The most important next step is to create a clearer backend map that agents can help maintain: systems, data flows, owners, permissions, failure points, and operational rules. This is the foundation for scaling the rest of the agent work.
Push PO Allocation and Cornelia forward
PO Allocation and Cornelia are central because they connect backend truth, customer communication, and post-purchase experience. They directly affect whether Lifely can create a reliable customer experience from order to delivery.
Protect marketing from operational friction
Marketing can generate growth, but that growth becomes fragile if the customer experience after purchase is confusing, delayed, or unsupported by clear systems. Q3 should connect marketing momentum with a stronger operational backend so demand does not outpace the business’s ability to serve customers well.
Keep Maurice moving, but treat it as a secondary operating track
Maurice and the HR system remain important, especially for hiring quality and process discipline. From a business-impact perspective, however, the immediate priority is lower than the operational backend, PO Allocation, and Cornelia.
Continue one-on-one coaching and skill building
The team is starting to build and use skills more effectively. Q3 should preserve the one-on-one coaching pattern where it is clearly producing results, especially when it helps team members create their own systems instead of depending on Stephane for every workflow.
Create a centralized business update rhythm
A lightweight meeting-summary or internal-newsletter system could help keep agent work connected to what is happening in the business. The goal is not more meetings. The goal is better shared visibility across marketing, operations, customer experience, and agent infrastructure.