Build1 distinct publisher3 min readUpdated
Foundry, announced August 17th, lets brokers describe a workflow in plain English and deploy it to one lane. The buy decision moves from which agents ship to who owns the process logic.
The Engineer · Build desk

Compiled by The EngineerSomething wrong?How this is made
Alvys said on August 17th that its freight transportation management system will include Foundry, a layer where carriers and brokers can deploy prebuilt AI agents, assemble their own by describing a workflow in plain English, or have Alvys engineers build one for them [1]. The third path matters less than the first two: a vertical software vendor is now selling the construction surface, which turns the buying question away from which agents come in the box and toward who owns the process logic once they are running.
The mechanics are ordinary software mechanics, and that is the point. Alvys says Foundry ships with more than 20 templates covering detention filing, document intake, rate auditing, track-and-trace, claims, compliance and load creation [2]. A user describes a workflow, reviews the generated process, tests it against simulated data, then deploys it to a selected lane or a larger book of freight [3]. Test environment, staged rollout, controlled blast radius: that is the shape of a deployment tool rather than a feature.
The agents act on the load, customer, document and financial data already held in Alvys [4], and the company's own framing concedes that autonomous actions inside a TMS can touch invoices, compliance records, customer communications and payments [5]. Hence Agent Shield, the governance layer. According to Alvys, administrators can set spending and approval thresholds, require human sign-off for consequential actions, pause agents and inspect an audit trail of decisions and overrides [6]; Alvys also says its enterprise agreements prevent model providers from training on customer data [7]. A model router assigns work by cost, speed and quality instead of routing everything to one large language model [8], the logic being that reading a standard proof of delivery is cheap work while an unusual claims dispute may warrant a premium reasoning model [9].
The rollout is narrow. Alvys previewed Foundry at its June 17th customer advisory board meeting, filled the initial customer cohort and opened a waitlist for a second [10], which puts about two months between private preview and public announcement [11]. The publisher's read is that this is closer to a staged production test than a broad release to every Alvys user [12].
Embedded agents in freight systems of record are no longer scarce. Trimble announced agents for order intake, invoice scanning and roadside breakdowns in November 2025 [13]. Tai TMS introduced an automated track-and-trace agent in April 2026 [14]. Transfix previewed AI exception management in May [15]. McLeod Software and Augment said on August 5th that Augment's agent would operate inside McLeod's PowerBroker TMS for carrier selection and shipment tracking [16], twelve days before the Alvys announcement [17]. Each of those is a named agent for a named task, shipped by the vendor. Runtimewire's assessment is that this activity blunts the claim that Foundry is built in rather than bolted on, and that the sharper distinction is the combination of a builder, reusable templates and implementation help inside one product [18].
The provenance is operator-side. Darman started the asset-based brokerage Archerhub in 2014 and assembled an engineering group after existing TMS products failed to fit its workflows; Alvys says that internal system helped Archerhub pass $90 million in annual revenue before it became a standalone product in 2020 [19]. Co-founder and CTO Leo Gorodinski was Jet.com's first employee and a vice president of engineering there, and Walmart acquired Jet.com for $3.3 billion in 2016 [20].
What to watch: whether the second cohort opens or stays gated, whether Agent Shield's audit trail starts appearing in customer procurement checklists, and who takes the call when a plain-English workflow written by a dispatcher files detention claims incorrectly. A builder moves maintenance of the logic onto the buyer, and brokerages are not usually staffed for that.
Follow any of these and your For You feed starts watching them — no settings page required.
Ranked by verification strength, evidence, and original report placement.
Alvys announced on August 17th that Alvys Foundry will let customers deploy prebuilt agents, assemble their own in plain English, or have Alvys engineers create them.
Alvys says Foundry includes more than 20 templates covering detention filing, document intake, rate auditing, track-and-trace, claims, compliance and load creation.
Users can describe a workflow in plain English, review the generated process, test it against simulated data and then deploy it on a selected lane or a larger book of freight.
Foundry adds an agent layer that can carry out multi-step operating procedures using the load, customer, document and financial data already stored in Alvys.
Autonomous actions inside a TMS can touch invoices, compliance records, customer communications and payments.
According to Alvys, administrators using the Agent Shield governance layer can set spending and approval thresholds, require human sign-off for consequential actions, pause agents and inspect an audit trail of decisions and overrides.
Evidence-backed comparisons of source perspectives and observed adoption signals. Read the methodology
Which Builder, Operator, and Investor concerns the observed source mix emphasized—not a truth score.
Evidence, demonstrated adoption, hype gap, incentives, and confidence are assessed independently, each on its own current evidence. How these are measured.
Single outlet, vendor-sourced capability claims
Every substantive detail traces to one trade-technology article reporting an Alvys announcement, with the most consequential specifics — template count, Agent Shield enforcement, the no-training contract terms — explicitly attributed to the vendor. There are no benchmarks, no named customers, no pricing and no third-party corroboration; the competitor timeline is the only externally checkable material, and the funding context is truncated mid-sentence.
One filled pilot cohort, waitlist beyond it
Observable adoption is a June 17th advisory-board preview, a first customer cohort described as filled and a waitlist for a second group — no counts, names, load volumes or outcomes. The article itself calls the deployment closer to a staged production test than a broad release. Sector-level adoption is better evidenced than Foundry's own: four rival vendors shipped or previewed embedded agents in the preceding ten months.
Infrastructure framing ahead of pilot-stage proof
Foundry is positioned as operating infrastructure with autonomous action over invoices, compliance and payments, while the demonstrated footprint is one filled pilot cohort and unverified vendor governance claims — that is overstatement, but modest, because the reporting supplies its own counterweights: it labels the rollout a staged test, argues embedded AI is becoming table stakes among rivals, and notes Foundry deepens rather than reduces vendor dependence.
Vendor launch narrative with founder framing
The story originates in a vendor announcement and quotes the founder's own grievance about paying for every customization, converting it into product positioning; capability, safety and template facts all come from the seller. Founder-credential material — Archerhub's $90 million revenue, a Jet.com acquisition figure — serves credibility building. The publisher partially offsets this by naming the lock-in consequence and the competitive context, but no counterparty, customer or competitor was given voice.
Consistent single account, thin verification
The narrative is internally coherent, dated and specific, and the competitive timeline is checkable, which supports moderate confidence in what was announced. Confidence in what Foundry actually does in production is low: one publisher, no independent testing, no customer evidence, no disclosed pricing, and two competitor dates given only to month precision.
invest
Your Landed Cost Is Being Litigated By Companies With $306,000 Problems1 distinct publisher
build
Invoked in three runs, executed in none: the cost rule that never got asked1 distinct publisher
invest
Klarna's $1bn Quarter: Margin Dollars Up 42%, Revenue Up 27%, And That Gap Is The Story2 distinct publishers
invest
The 30-Year Sets a Record and Everything Long-Dated Gets Repriced1 distinct publisher
Distinct publishers with included, body-backed reporting in this cluster.
1 article · August 17, 2026