10 October 2026 · Samuel Mirpuri
AI agent development services: what you're buying, and how to tell a working agent from a prototype
AI agent development services can produce a convincing prototype in days. What a buyer pays for is everything after it: integration, evaluation on real data, exception handling, oversight in production and someone accountable for the result. What to ask for, and what the evidence says about agent projects that stall.
Read the essay- 8 Oct 2026
Javier LeguinaProcure to pay: the process step by step, and where automation pays
The procure to pay process runs from a request to buy something to a recorded payment. What each step does, how it differs from source to pay, and why most of its cost sits in invoice exceptions rather than in the buying. - 6 Oct 2026Samuel Mirpuri
Process improvement consulting: what it delivers, and why the gains fade
Process improvement consulting brings Lean, Six Sigma or reengineering to a business, and the savings during the engagement are usually real. The evidence on what happens after the consultants leave is less flattering. What the work delivers, why the gains fade, and what to ask for instead. - 3 Oct 2026Samuel Mirpuri
ERP migration: what it costs, why it overruns, and the alternative
An ERP migration is the most expensive way to fix a process problem and the most common. What a migration involves, what Panorama and Gartner say about its odds, the two failures every CFO knows, and when a layer on top of the existing system is the better answer. - 1 Oct 2026Samuel Mirpuri
Business process automation consulting: what to expect, and what to demand
Business process automation consulting comes in three shapes: a roadmap, a bot on a license, or a working agent that someone runs. Here is what each delivers, what the engagement should contain, and what to write into the contract. - 29 Sept 2026Samuel Mirpuri
flowscope raises $3.8M seed round
flowscope has raised a $3.8M seed round, backed by Y Combinator, Red Forest Capital, Anderson Growth Partners and Outbound Capital. - 29 Sept 2026Samuel Mirpuri
AI in private equity: where the value is in portfolio operations
The deals in today's funds need 10 to 12 percent annual EBITDA growth to return what 5 percent used to. AI in private equity is being asked to find that growth, and the evidence now says where in a portfolio company it pays. - 23 Sept 2026Javier Leguina
Rebuilding cash application: matching the payment to the invoice when the remittance arrives separately
Cash application is slow because the remittance information that explains a payment usually arrives apart from the money. We walk the redesigned workflow, what it mechanizes, and where the analyst keeps the call. - 21 Sept 2026Samuel Mirpuri
Why a frontier-model company is not the natural home for this work
The large AI labs are building forward-deployed services arms in 2026, and the structure of those arms is the answer to whether they should do this work themselves. - 18 Sept 2026Javier Leguina
The unit economics of the human-in-the-loop exception tail
In a production automation, cost is set by the small fraction of cases that route to a person, not by the volume the system handles cleanly. Here is how to quantify that tail from the same benchmarks that describe each process. - 16 Sept 2026Samuel Mirpuri
How enterprise AI procurement is changing
Buyers are moving from evaluating advertised AI capability to validating production results on their own data. The shift is a rational response to the gap between AI spending and results, and it favors providers that can show a running result over a demonstration. - 14 Sept 2026Samuel Mirpuri
Process redesign inside a freight broker or third-party logistics provider
A freight broker is one instance of the horizontal back-office pattern, in an industry where thin margin makes every transaction-level touch consequential. Here is what the redesigned back office actually does, load by load. - 11 Sept 2026Samuel Mirpuri
How a forward-deployed engineering organization is actually built and staffed
The hard part of delivering working AI on a client's systems is staffing a role that combines production engineering, client elicitation, and judgment about exceptions, and that combination is what limits the model's scale. - 9 Sept 2026Javier Leguina
The objection that people will resist this, addressed to leadership
Workforce resistance is the most common reason leaders give for why an automation effort will stall, and the fear underneath it is real and measurable. The most-quoted statistic about change failure is not. Here is what the reliable research actually supports, and the design choice that answers the fear instead of confirming it. - 7 Sept 2026Javier Leguina
Rebuilding sales-tax compliance and filing
Sales-tax compliance grew structurally after one Supreme Court decision and is still done largely by hand. A worked example of the redesign, and why the filing calendar, not rate lookup, drives the labor. - 4 Sept 2026Samuel Mirpuri
Process redesign inside an accounting firm
Inside an accounting firm, the binding constraint is people, not budget. That turns automation into a capacity question, and the redesigned workflow is how a firm takes on more work without hiring people it cannot find. - 2 Sept 2026Samuel Mirpuri
The consulting industry under AI, in the data of its own restructuring
The repricing of consulting is no longer speculative. The 2025-26 layoffs and the slow move toward outcome-based fees expose a mechanism: the major firms sell human hours, and AI reduces the hours each piece of work requires. - 31 Aug 2026Javier Leguina
Rebuilding expense and travel processing: the cost is in the error tail
Most of the cost of expense reimbursement sits in the fraction of reports that come back wrong. A redesign that captures receipts, checks them against policy, and routes only genuine exceptions to a person recovers that time without touching finance approval. - 28 Aug 2026Javier Leguina
Idempotency and exactly-once guarantees when an agent writes into a system of record
When an agent posts an invoice or issues a payment, retries make duplicate writes a near-certainty unless the writes are designed to be safe to repeat. Here is the discipline that makes them correct. - 26 Aug 2026Javier Leguina
We tried RPA and it failed, and why agents are a different mechanism
Operators who burned a budget on robotic process automation are right to be skeptical of the next automation pitch. The honest answer is to explain why a model-based agent fails differently, not to insist it cannot fail at all. - 24 Aug 2026Javier Leguina
Rebuilding payroll: validating the run before it goes out
Payroll is a high-frequency process where manual data assembly produces a recurring, measurable error rate with compliance and retention costs. Here is the redesigned validation workflow and what it catches before the run leaves. - 21 Aug 2026Samuel Mirpuri
Process redesign inside an IT managed services provider
An MSP runs on technicians clearing tickets, alerts, and reports, and its margin tracks how much each one can carry. Here is the redesigned internal workflow, and why the buyers rolling these firms up care. - 19 Aug 2026Javier Leguina
Prompt injection when an agent reads untrusted documents and emails
An agent that reads an inbound email or a shared document can be made to obey instructions hidden inside it, because the model treats trusted commands and untrusted data as one stream. Here is why the failure is architectural, and the containment that holds. - 17 Aug 2026Javier Leguina
Rebuilding the back office of customer support: triage, routing, and resolution
The cost of customer support hides behind the chatbot, in the internal work of classifying, routing, and resolving a ticket. Here is what that work costs and how the redesign reclaims it. - 14 Aug 2026Samuel Mirpuri
Build, buy, or hire an AI capability: the decision an operator actually faces
Before you evaluate any AI vendor, you face a prior decision: build the capability in-house, buy a working result, or hire for it. The choice turns on three quantities you can estimate. - 12 Aug 2026Samuel Mirpuri
The AI adoption gap by company size, and the mid-market's advantage in time to production
Adoption rises with firm size, not the reverse. The mid-market starts behind on the headline rate and can close it fast, because the path from decision to a working deployment is short. - 10 Aug 2026Javier Leguina
Controls for autonomous actions: audit trails, reversibility, and human oversight
When an agent posts a journal entry or releases a payment, the question that matters is not average model accuracy but whether each action is logged, attributable, reversible, and overseen. Three named frameworks now say what to control, and they map cleanly onto engineering. - 7 Aug 2026Samuel Mirpuri
Process redesign inside a wholesale distributor
In distribution, the labor sits in order entry and pricing exceptions, not the catalog. Walking order-to-cash at the line level shows why transaction efficiency decides where thin margins land. - 5 Aug 2026Javier Leguina
The objection that the data is too messy, and why it argues for a different method, not against automation
Messy data is real and expensive, and it defeats rules that require clean input. That is exactly why the case for automation rests on a method that reads the data as it is and routes the doubtful cases to a person. - 3 Aug 2026Javier Leguina
Onboarding a customer or vendor without the manual file
Onboarding a counterparty is slow because the same documents get requested and rekeyed across systems, not because the verification rules are hard. Here is the redesigned workflow, with the decision kept human. - 31 Jul 2026Samuel Mirpuri
What the operator's role becomes after the manual work is automated
When an assistant compresses the experience curve, the residual human work moves toward judgment, exceptions, and oversight. The labor economics says this is measurable, and it tells you who to hire and how to train. - 29 Jul 2026Javier Leguina
The regulatory trajectory for enterprise AI agents
Compliance work has a lead time, so an operator deploying agents needs the direction and timing of the rules, not just today's text. Here is the trajectory, and why auditability designed in early is cheaper than auditability retrofitted. - 27 Jul 2026Samuel Mirpuri
Process redesign inside an insurance brokerage
A brokerage controls its margin through administration expense, not commission, and that expense concentrates at renewal. We walk the renewal workflow and show what an agent does and what stays with a licensed broker. - 24 Jul 2026Javier Leguina
Three-way match, rebuilt: matching the order, the receipt, and the invoice before payment
The control that prevents overpayment is also the largest source of manual work in accounts payable. The redesign turns the matching step into the easy part and moves the real work upstream. - 22 Jul 2026Javier Leguina
The hallucination objection, answered
"Models hallucinate" is correct in the abstract and misleading as a blanket disqualifier. The rate depends on the model, the task, and whether the output is grounded in source text and verified, which is what turns it into a measured, managed quantity. - 20 Jul 2026Javier Leguina
Monitoring an agent after it ships: drift, regression, and evaluation in production
A deployed agent can change behavior with no change on your side, because the model provider updated something and did not version it. Here is the production machinery that catches it before your customers do. - 17 Jul 2026Samuel Mirpuri
Process redesign in healthcare revenue-cycle management
Revenue-cycle management is one organization moving claim data between systems, and the denial benchmarks are public enough to redesign the work against. Here is the rebuilt workflow, and why the recoverable share of a denied claim falls the longer it sits. - 15 Jul 2026Samuel Mirpuri
What changed in model cost that makes this work now
The make-or-buy math on automating a document-heavy process flipped because the unit cost of running a model collapsed, not because any one capability arrived. Here is what a non-technical operator should take from the cost curve. - 13 Jul 2026Javier Leguina
Working a receivables ledger down: a redesigned collections process
Late payment is a working-capital cost most operators carry without measuring it. We walk one redesigned collections workflow and the mechanism by which it closes the gap between what is invoiced and what is collected on time. - 10 Jul 2026Samuel Mirpuri
Access is the moat, not the algorithm
When any competitor can generate equivalent agent code, the scarce asset is permission to write into a customer's production systems and the accountability that earns it. Why outcome-aligned pricing and embedded delivery are the structures that win and keep that access. - 8 Jul 2026Samuel Mirpuri
The context flywheel: how each engagement makes the next one cheaper and the agents better
In AI services, code is becoming copyable and cheap, so it cannot compound into a moat. What compounds is accumulated context: the library of exceptions, document variants, and write-back patterns that each deployment teaches the shared agent infrastructure. This post states that loop literally and argues why it is a context flywheel a new entrant cannot buy. - 6 Jul 2026Samuel Mirpuri
Where value accrues when the code is free: the cost anatomy of an AI services engagement
When model inference and code generation cost almost nothing, the dominant terms in an AI services engagement become context acquisition and the human-in-the-loop exception tail. A line-by-line walk of which costs fall toward zero and which stay. - 3 Jul 2026Samuel Mirpuri
Why an agent does discovery and delivery better than a consultant does either
A consulting engagement loses information at every handoff between discovery, redesign, build, deployment, and monitoring. One agent holding the whole arc removes the self-report gap and the handoffs at once, which is where the accuracy and the speed come from. - 1 Jul 2026Samuel Mirpuri
Forward-deployed agents, not just forward-deployed engineers
The third-layer story has centered the scarce human forward-deployed engineer. The next move is that agents do most of that work, observing, mapping, building, and monitoring, while a small human core keeps the judgment and accountability that model capability relaxes but does not remove. - 29 Jun 2026Samuel Mirpuri
Code was never the moat: context is the only durable asset in AI services
The deliverable a services-as-software firm sells and the asset that makes it defensible are two different things. As model inference collapses in price, the code an agent runs commoditizes, and the durable moat moves to the observed, accumulated context of how a specific business actually runs. - 28 Jun 2026Samuel Mirpuri
A map of who else sits in the third layer
The forward-deployed services model has gone from a Palantir curiosity to a contested category with labs, vertical startups, and services firms all in it. Here is who's in it, and where an independent mid-market implementer sits. - 26 Jun 2026Samuel Mirpuri
How to baseline a process well enough to stand behind a result
An outcome-aligned engagement is only as honest as its baseline, and most baselines are self-reported and wrong. Here is the measurement machinery that makes a before-and-after comparison defensible. - 24 Jun 2026Javier Leguina
What process redesign looks like inside an industrial staffing firm
A workflow-level walk through recruiting and onboarding at a staffing firm, where the delays in time-to-submit and time-to-fill come from, what an agent can run, and where a human stays in the loop. - 22 Jun 2026Samuel Mirpuri
When not to automate, and what an honest engagement does when the savings don't show up
A clear-eyed account of the workflows that make bad automation candidates, why first principles predict it, and what a vendor actually does when the projected hours never appear. - 20 Jun 2026Javier Leguina
The employee-monitoring laws that decide how an observation agent can be deployed
A handful of facts about US law govern where a capture agent can run, and the bill most blogs called a 2026 law never passed. This is what's actually on the books, and why flowscope adopts the failed bill's principles anyway. - 18 Jun 2026Javier Leguina
The long tail of document variability is the whole job
"Automate data entry" names only the easy part of the work. Rules and template OCR leave a stubborn share of every document stack for humans to key, and the long tail of document variability is the whole job, where a language model plus a human earn their place. - 16 Jun 2026Javier Leguina
What the agent-reliability curve says about which workflows are automatable now
A mid-2026 reading of the METR task-length curve, turned into a workflow-selection rule for operators deciding what to automate this quarter and what to wait on. - 14 Jun 2026Javier Leguina
Writing back into a system that has no usable API
Most mid-market automation stalls at the integration surface, not the model. The engineering case for acting at the desktop a clerk already uses, walked through the accounts-payable queue and QuickBooks. - 12 Jun 2026Javier Leguina
Shadowing instead of surveillance, and why the difference is behavioral, not cosmetic
The objection to an observation agent is grounded in real evidence: monitoring backfires when its data disciplines individuals. Diagnostic shadowing inverts every variable the research blames. - 10 Jun 2026Javier Leguina
Rebuilding the month-end close, from trial balance to statements
The monthly close is the cleanest worked example in operations because the benchmarks are public. We walk one redesigned close end to end, from where the days actually go to what a finance team does with them back. - 8 Jun 2026Javier Leguina
What a capture agent records, and what it's built to throw away
The first thing operators ask about an observation agent is what it collects and where the data goes. We answer with the actual architecture: scoped capture, redaction at the endpoint and again before storage, minimum retention, and terms you can check at trust.flowscope.com. - 6 Jun 2026Javier Leguina
Why documented processes rot, and why watching the work beats reading the SOP
Standard operating procedures go stale because the real process lives in tacit knowledge that resists being written down. Observing the work recovers the layer documentation cannot. - 4 Jun 2026Javier Leguina
How production reliability gets engineered, and why a demo is not evidence of it
A pilot at eighty percent on a clean slice tells you almost nothing about whether the workflow runs unattended on Monday. The machinery that closes the gap, with named benchmark numbers. - 2 Jun 2026Samuel Mirpuri
How to tell an AI delivery vendor that ships from one that demos
A buyer's procedure for separating vendors that put working software on your data from vendors that sell decks, built on the base rate that most organizations have not yet scaled AI beyond pilots. - 31 May 2026Samuel Mirpuri
Why services-as-software firms scale where AI consulting cannot
A services firm that ships software, prices against outcomes, and reuses agents across customers is not the same business as a consulting firm with AI bolted on. The unit economics diverge from inception. - 29 May 2026Samuel Mirpuri
What an aligned AI engagement actually looks like
An engineer in the customer's environment from day one. Discovery in days. Redesign before automation. Pricing against measured savings. This is what an aligned AI engagement looks like end to end. - 27 May 2026Samuel Mirpuri
There's a third layer of consulting that nobody has named yet
The consulting market has had two layers for fifty years. A third has been forming as AI spreads: teams that wire AI into live systems, govern them in production, and stay accountable six months after the platform vendor moves on. - 25 May 2026Samuel Mirpuri
The 80-to-99% problem
Foundation Capital named the ratio: eighty percent of capability with twenty percent of effort gets you to a pilot, and the remaining nineteen percent requires roughly one hundred times more work. That work is engineering, in the customer's environment. - 23 May 2026Samuel Mirpuri
The unbillable hour
Professional services firms still selling time are watching their revenue base compress into the technology. The unbillable hour is the structural P&L problem at the center of every consulting firm in 2026, and outcome pricing is the only way out. - 21 May 2026Samuel Mirpuri
Services-as-software is the right frame. AI roll-ups are the wrong one.
Two competing strategies are bidding for the post-SaaS opportunity. They are not the same bet. The roll-up plays for multiple arbitrage. Services-as-software builds the firm with software economics from inception. - 19 May 2026Samuel Mirpuri
The length of the discovery phase is a business-model decision
Interview-based process discovery produces an authored map that starts going stale the day it ships. Observation-based agents produce a more accurate map in days and keep it current. The duration of discovery is set by the engagement model rather than by the work. - 17 May 2026Samuel Mirpuri
The customer all four delivery models leave behind
There are four ways to buy enterprise AI in 2026. Each assumes a customer profile. Each leaves the same business behind: the mid-market operator with no AI team and a manual workflow consuming a meaningful share of payroll. - 15 May 2026Samuel Mirpuri
Don't automate. Obliterate.
Michael Hammer's 1990 essay is more correct in 2026 than it was then. AI is automating cow paths instead of obliterating them, and the seventy-five-percent reduction nobody captures is the cost. - 13 May 2026Samuel Mirpuri
Why your AI consultants left you a deck
By month six of a typical Big-4 AI engagement, the deliverable is a deck and a pilot proposal, with nothing written to a system of record. The model serves the customer it was built for. Most operators are not that customer. - 11 May 2026Samuel Mirpuri
Where this goes next: the system of action
Systems of record are not going anywhere. What changes is the layer above them: the system of action that reads, acts, writes, and stays accountable for the result. - 9 May 2026Samuel Mirpuri
The state of enterprise AI in 2026: a map
Five delivery models are bidding for enterprise AI budgets in 2026: Big-4 transformation, AI boutiques, SaaS copilots, in-house teams, and AI roll-ups. None serves the largest segment. - 7 May 2026Samuel Mirpuri
SaaS is dead: long live services-as-software
Two trillion dollars came off software stocks, the steepest non-recession drawdown in decades. The repricing names what comes after SaaS: services-as-software, where the customer buys the outcome and AI delivers it.