Points of view
POV
Short, opinionated takes from the Head of AI seat. Four streams: retail and CPG for the C-suite, B2B GTM for operators, AI and Agentic for builders shipping production agents, and Agentic Retail for retail and CPG boards deciding what to fund, govern and stop.
Perishables are where your forecasting model gets graded
Center-store errors hide in inventory. Fresh errors show up as waste in days. Perishables are the only honest scoreboard for a demand model.
AI governance for teams that don't have a CISO yet
Most teams ship agents with shared API keys and no logging. Here is the minimum governance layer to build before you have a security org.
Every retailer is building the same three agents, and that is the tell
When a whole sector's agent roadmap looks identical, it is vendor-led, not strategy-led. Here is how to tell whose roadmap you are actually funding.
Every POS upgrade is an AI decision now
Retail scopes POS refreshes as seven-year hardware buys. But the terminal is the data-capture layer every agentic workload depends on. Buy it wrong and you cap your ceiling.
CODN for B2B SaaS, a worked example
Most GTM leaders ask what an agentic rebuild costs to build. The sharper question is what it costs to do nothing. Here is the full CODN math for a Series C SaaS.
The agent failure mode taxonomy nobody publishes
The demo-friendly failure list is wrong. The four categories that actually kill production agents are stale data, missing context, hallucination, and conflict-with-truth.
Measure agents on resolution rate, not deflection
Deflection is a vanity metric that buries downstream cost. Resolution is harder to game, and it is the only number that pays the bill.
Eval harnesses that catch calibration drift
Your agent's demo score is not lying. Your harness is. Here is how to design evals that surface the calibration drift quietly killing production agents.
The agent is the new shopper and your product data is written for humans
When an AI agent mediates the purchase, product data becomes the interface. Structured, machine-readable catalogs are now a revenue line, not a technical chore.
Assortment planning still runs on last year shopper
Assortment decisions describe a customer who already changed. Here is where to inject live signal without turning merchandising into a slot machine.
What "AI-native" actually means when you're hiring
Every B2B job post now says 'AI-native.' The data says it's mostly positioning. Here is the operational definition that separates signal from buzzword.
The three-day MCP server design playbook
A senior operator's three-day sequence for shipping an MCP server that compounds instead of decaying into an API wrapper nobody wants to call twice.
Your vendors are repricing from seats to outcomes and procurement is not ready
Retail vendors are shifting from seat licenses to per-outcome contracts. Procurement built to score seats cannot score outcomes, and that gap is where margin leaks.
Your supply chain twin is a slideshow, not a simulation
Most retail digital twins visualize the past. A twin worth the spend runs counterfactuals fast enough to change tomorrow morning.
AI SDRs did not fail, your offer did
The AI SDR post-mortems blame the tooling. The real failure was sending a weak offer faster, to more people. Automation amplifies whatever was already there.
Multi-agent is a coordination tax you may not need
Splitting a working single agent into a swarm usually buys an org-chart metaphor and pays in latency, failure surface, and debugging. When the tax is worth it, and when it is cosplay.
Human in the loop is a budget line, not a principle
Everyone claims human oversight for retail agents. Almost nobody funds the reviewer hours it requires. Here is how to size it and price the Cost of Doing Nothing.
The clean room is where first-party data goes to die
Clean rooms got sold as the privacy-safe fix for signal loss. In production they add latency, cap match rates, and starve the real-time decisions agentic retail actually needs.
Your agent needs a budget, not just a prompt
Agents fail open on cost. No ceiling, no step limit, no breaker. Treat spend as a design constraint, not a dashboard you check after the invoice lands.
Your data is not ready, and the readiness project is the trap
Enterprise-wide data readiness programs defer value past the patience of the people funding them. Scope readiness to one workload at a time.
Shrink is a data problem wearing a security costume
Shrink lives with loss prevention, but the variance sits in inventory accuracy, markdown timing, and POS exceptions. It reports to the wrong function.
The CODN of a slow sales cycle
Everyone models revenue from the win. Almost nobody models the compounding cost of ninety extra days to get there. That is where mid-market GTM value leaks.
The model is not your bottleneck, your context is
Teams keep upgrading models to fix broken agents. The real constraint is what the model can see. Here is where the leverage actually sits.
Agent governance is an access-control problem wearing an ethics costume
Most enterprise agent governance produces principles decks. The real risk surface is what an agent can read, write, and spend. Three questions expose whether your program is real.
Dynamic pricing without trust is churn with extra steps
Price optimization can win the transaction and lose the customer. The guardrails that separate durable dynamic pricing from the version shoppers punish.
Churn is a sales problem you booked twelve months ago
Retention teams inherit deals that were mis-sold, then get graded on saving them. Trace churn back to qualification and the CODN math changes.
MCP is a distribution play, not a protocol
The MCP spec is the least interesting part. The real fight is over who owns the tool surface agents reach for by default.
You do not have an AI strategy problem, you have an ownership problem
AI now spans merchandising, marketing and supply chain with nobody accountable for the tradeoff between them. The operating model decides outcomes, not the model.
Returns are a pricing decision you keep making by accident
Your return rate is not a logistics line. It is a signal about price, fit, and copy. Managing it as reverse logistics guarantees you keep paying for a merchandising problem.
Your pipeline review is a fiction-writing workshop
Weekly pipeline reviews reward storytelling over evidence. Three questions that turn the meeting from theater into a forecast you can bank.
The eval you skip is the outage you schedule
Every agent incident traces to a failure mode someone called too rare to test. Here are the three evals that cover the majority of your real risk surface.
Pilots do not scale because nobody owns the exception queue
Agents nail the happy path. The exceptions land on a team nobody staffed, and that is where the pilot quietly dies. Exception ownership is the scaling constraint.
The Cost of Doing Nothing is the only AI number your board will believe
Boards discount every vendor ROI projection. CODN reframes the question from what you might gain to what you are already losing, measured from data you own.
Your labor model is the last analog system in the store
Retailers digitized inventory, pricing, and marketing. Store labor still runs on a spreadsheet and a manager's gut. That gap is where service, shrink, and conversion get decided.
You do not have a lead problem, you have a routing problem
Teams patch a pipeline gap by buying more leads. The CODN math almost always shows the loss sitting in routing latency and ownership ambiguity, which costs nothing to fix.
Most agent demos are just a for-loop with anxiety
A retry loop around one prompt is not an agent. State, tool choice, and evals are the line between a demo and production.
Your SDR team is a symptom, not a strategy
Adding SDR headcount when routing, qualification, and follow-up are broken is not a strategy. It is a bill. Here is the diagnostic that tells you which problem you actually have.
Retail media is eating the P&L, and nobody owns it
Retail media now moves enough margin to distort merchandising and pricing, yet it sits inside marketing with no cross-functional owner. That gap is the risk.
The store associate is your best AI training data
Retailers are licensing corpora and buying synthetic data while the highest-signal operational data in the business walks the floor every day, uncaptured.
Your loyalty program is a liability, not an asset
Unredeemed points, stale consent, and no agent-readable profile layer turn loyalty into a balance-sheet obligation and a privacy exposure. Here is the CODN math on rebuild versus retire.
ICP isn't a doc, it's a system
Your ICP doc is dead the moment you save it. The teams winning pipeline in 2026 run ICP as a living system that learns from every won and lost deal.
Outbound theater versus outbound
Most outbound in 2026 is theater: sequences fire, replies trickle in, calendars stay empty. Three signals that tell the difference between the show and the system.
CODN for retail, a worked example
A full Cost of Doing Nothing walk-through for a Tier 2 specialty retailer weighing agentic merchandising, with real numbers and a real decision.
What Walmart's ad business actually teaches
The lesson is not retail media. It is what happens when first-party data becomes a product instead of a byproduct, and why most retailers are copying the wrong layer.
Custom MCP servers that compound vs ones that don't
Five practices that separate production-grade MCP servers from glorified internal scripts, written from the Forward Deployed seat.
Why "agentic" should mean operates, not responds
Production agents do not look like chat. They look like distributed systems. The operator pattern beats the chatbot pattern, and the CODN on getting this wrong is brutal.
The agent tax: hidden costs of bad MCP servers
Script-grade MCP servers feel free in the demo and bill you for years in production. Here is what they actually cost, and how to skip the retrofit.
The retail board memo every CIO should write this quarter
Capital markets reward AI announcements, but the P&L lags. Here is the CODN-framed board memo retail CIOs should write before the 2026 budget locks.
The CODN audit every CRO should run before Q4
A 90-minute Cost of Doing Nothing audit a CRO can run with their RevOps lead before Q4 budget conversations. No slide deck. No 30-day project.
Demand forecasting has a signal problem, not a model problem
Retailers keep buying smarter forecasting models to compensate for dirty inputs. The bottleneck is signal quality, not algorithms.
AI governance is a logging problem before it's a policy problem
Enterprises write AI policy before they can observe what their agents do. You cannot govern what you do not log. Instrumentation comes first.
Your seven-tool GTM stack is one tool too many
Most GTM stacks carry one redundant tool that quietly drains budget, attention, and data quality. Finding and cutting it is the highest-leverage RevOps move of 2026.
CPG trade promotion is retail's most underused AI workload
CPGs spend up to a quarter of revenue on trade promotion and 72 percent of it loses money. That is the AI workload nobody is prioritizing.
Context engineering is the job prompt engineering pretended to be
Prompt wording barely moves the needle in production. Context assembly does. Here is what actually breaks when you treat the window as a scarce resource.
Intent data without orchestration is expensive noise
Most B2B teams buy intent data, park it on a dashboard, and miss the conversion window. The value was never the signal. It is the orchestration that acts on it.
The omnichannel measurement stack that actually closes the loop
Most omnichannel measurement stacks report but never act. Closing the loop needs identity, incrementality, and an action layer that moves spend.
Your agent doesn't need more tools. It needs fewer.
Tool sprawl is the silent killer of agent reliability. Cutting tools, not adding them, is what makes agents production-grade.
The agentic outbound stack that beats the agency model
Agencies bill 15K to 30K a month for outbound a 2K agentic stack now runs. The math flipped. Here is what the stack contains and where agencies still matter.
Personalization is back, and worse than before
The personalization vendors that died in 2023 are reborn as agentic. Most of it is the same broken assumptions in a better wrapper.
The CFO's AI question every CDO is failing
There's one question retail CFOs are starting to ask their CDOs about AI investment. Most CDOs aren't ready for it — and the ones who are change the conversation from ROI to CODN.
Your scoring model is the wrong fight
Every GTM team is debating signals, weights, fit vs intent, predictive vs explainable. They're fighting the wrong fight. The model rarely matters. The orchestration around it does.
Evals, or it's vibes
Every production AI agent needs an eval harness. Without one, you're shipping vibes — and you'll only find out when the agent acts on bad data at scale, in production, repeatedly.
The $40M data lake nobody asks about anymore
Three years ago every Tier 1 retailer built a data lake. The AI conversation moved on and the lake went quiet. Here's the retrofit the winning retailers are quietly running.
MCP is the protocol the agent conversation needed
Most 'AI agents' fail not because the model is wrong but because the agent has no way to read your data and act on your systems. MCP is the missing layer. Here's why it's the unlock.
RevOps is becoming a build job, not a tools job
Five years ago RevOps was a tools-admin role. In 2026 it's a build role. Leaders are hiring engineers who happen to know GTM — and the middle market is one cycle behind.
Why Claude Code is the RevOps hire of 2026
RevOps is becoming a build job. Most teams can't hire fast enough at the new bar. Claude Code closes the gap — if you treat it like a junior engineer, not a chatbot.