Skip to content

San Francisco · Irvine · New York · Hong Kong

Bringing AI into real workflows

HZ Tech is a forward-deployed engineering (FDE) consultancy. Our engineers work inside your operation to find the problems worth solving, build AI into the workflow and turn what repeats into standard products. Our work spans logistics, manufacturing and trade.

  • Seed round closed
  • Products live with customers
  • International founding team
How we engage
An HZ Tech autonomous forklift robot in a warehouse aisle

What our engagements have delivered

Product figures from live deployments; partner figures from the customer's public information

7products built in FDE engagements, 2 of them live
20%+higher peak throughput at a live warehouse, with 30%+ lower cost
30+employees use our general office agent platform, including customers in other industries
~1,000people at the logistics group we partner with exclusively, across 30+ branches
01Industry Insight

Companies don't lack software — they lack connection

The real problem looks the same in every industry: software, data and people's work are cut off from one another.

Scattered requests

Requests arrive by email, chat, spreadsheets and phone, and no one sees the whole queue.

Inconsistent data

Rate sheets, price lists and specifications come in different formats and keep changing.

Work across portals

Bookings, filings and orders are spread over portals and partner systems — air-cargo bookings and customs filings are typical.

System silos

ERP, TMS, WMS, finance and quality systems don't talk to each other.

Duplicate entry

Staff copy, clean and re-key the same data between systems.

Operational black box

Management has to reprocess data before it can see how the business is doing.

Research findings

Three findings from the field.

Across current industry research and our live product work, the signal is consistent: value appears when data, workflow and action are designed as one operating system.

Exhibit 01Adoption is rising faster than realized value
~90%

of surveyed shippers had adopted at least one transportation AI use case

Logistics survey
96%

had deployed at least one surveyed AI or digital use case in warehousing

Logistics survey
8%

of surveyed manufacturers reported achieving all targets from AI initiatives

Manufacturing survey

Sources: McKinsey 2026 State of Digital Logistics Survey; BCG 2025 manufacturing survey. Different samples and questions; figures are directional and not directly comparable.

Finding 01Adoption is broad. Operational value is not.

Enterprise adoption is no longer the strongest signal of advantage. Recent logistics research shows digital tools moving into the operating core, while manufacturing research shows a wide gap between deploying use cases and achieving their intended targets.

More tools will not resolve a weak operating model. Value depends on whether teams can move from detection to decision and from decision to execution without losing context, ownership or control.

Finding 02The system of record is becoming a system of action.

Traditional software records what has already happened. The emerging operating layer must assemble the current context, make the next decision legible and route an approved action back into the workflow.

A supply chain cannot be coordinated if rates, bookings, documents and milestones disagree. A factory cannot be optimized if orders, batches, capacity and quality evidence are separated. The operating record is the substrate on which useful intelligence runs.

Finding 03The method repeats. The operating truth does not.

Logistics and manufacturing share a delivery discipline: observe the work, structure the record, embed intelligence, operate and improve. But they do not share the same ontology, event model or decision cadence.

Logistics follows the supply chain across organizational boundaries. Manufacturing follows an order and batch through controlled production stages. A generic workflow removes the context that makes the system valuable.

Sources & notesResearch base.

External research frames the market view. Live product repositories and operating documentation ground the functional case work.

  1. 01Goldman Sachs Research — How Industries Are Putting AI to Work (External link)Industry adoption, workflow-level agents and enterprise-data constraints
  2. 02McKinsey — Creating Value from AI and Digital Capabilities in Logistics Operations (External link)Adoption, operating decisions and connection to execution
  3. 03BCG — The New Cost Calculus for Manufacturers (External link)Manufacturing adoption and the gap between use cases and realized targets
  4. 04IATA ONE Record (External link)Shared cargo record and standardized supply-chain data exchange
  5. 05GS1 EPCIS (External link)Event visibility across what, when, where, why and how
  6. 06NIST — Artificial Intelligence for Manufacturing (External link)Production planning, integration, human collaboration and traceability
  7. 07HZ Tech product evidence — Logistics operating platform (External link)Orders, booking, customs, warehouse events and controlled execution paths
  8. 08HZ Tech product evidence — Rate intelligence (External link)Rate ingestion, normalized records, natural-language query and source-aware answers

Methodology note. External survey figures use different samples and definitions and should be read directionally. Product-scope figures describe implemented system structure; they are not customer outcome claims.

Report / 2026 · HZ Tech · Our insights8 minute read · Logistics + manufacturing · November 2026Read the full thesis, evidence and research notesWhen intelligence enters the workflow.Why the next operating advantage depends on structured context, controlled action and an industry-specific record of work.
02Timing

Why now

Three technology inflection points and one shift in awareness are converging on the same timeline.

  1. 01Cost inflection

    AI for everyone

    Since the “DeepSeek moment”, open-source models have matched closed models on many benchmarks and token costs have fallen sharply — enterprise AI went from unaffordable to affordable.

  2. 02Capability inflection

    Agents that work

    A new generation of models such as GPT-5.3 and Claude Opus 4.1 has matured long-horizon planning and tool use, taking agents from demos into real workflows.

  3. 03Ecosystem inflection

    Vertical deployment

    Models such as Claude Fable 5, GPT-5.6 and Kimi K3 make it realistic for agents to embed reliably into a vertical industry's systems, data and collaboration ecosystem.

  4. 04Awareness inflection

    Demand awakens

    With domestic models like Doubao and general-purpose agents like WorkBuddy now widespread, managers and frontline staff have used AI first-hand — the question has shifted from “should we use AI?” to “how do we bring AI into our own business?”

Executive summary · The HZ view

AI is moving from the interface to the operation.

The first wave of enterprise AI made information easier to find and content faster to create. The next wave is harder—and more valuable. It must understand the record of work, recognize the decision in front of an operator and connect that decision to a controlled action.

Read the full summary

A chatbot can sit above fragmented systems. An operating product cannot. In logistics, it needs the current supply-chain record, rate, document, milestone and partner status. In manufacturing, it needs the order, specification, batch, production stage and quality evidence. Without that context, intelligence may sound convincing while remaining operationally irrelevant.

The durable advantage will not come from adding intelligence to the edge of an old process. It will come from digitizing the work deeply enough that intelligence can participate in the process—reliably, traceably and under human control.

Key takeaways

  • Adoption is not the moat. The differentiator is how quickly insight becomes an operating decision.
  • Context is the missing layer. Intelligence needs a structured record of work, not another disconnected data feed.
  • The method can repeat. The workflow and data model must still be rebuilt around each industry.
The question has changed

Not “Can AI answer?” What happens next?

Technology is no longer the biggest barrier. The biggest barrier is now ecosystem integration and the ability to actually deploy intelligent operations inside enterprises.

Whoever has real scenarios, industry understanding and a delivery method stands on the right side of the divide. That is why HZ Tech chose the FDE path: we work inside our customers' operations, one industry at a time.

03Our Method

FDE goes into the field; products come out of it

Solve concrete customer problems like an FDE team; build repeatable standard products like a SaaS company.

  1. 01Go into the operation

    Interview frontline roles and observe real work

  2. 02Break down measurable problems

    Split into steps, inputs/outputs and acceptance criteria

  3. 03Build a usable prototype fast

    Validate the workflow quickly with AI-assisted development

  4. 04Embed in the real process

    Iterate on real data, keeping human approval and rollback

  5. 05Extract standard capabilities and replicate

    Consolidate into standard products — agentic systems, skills, MCPs — to cut deployment cost

The operating model

Build the context. Then close the loop.

A useful industrial system does not stop at insight. It connects three layers of work and makes the transition between them explicit.

01 / Record

Capture the work.

Structure the documents, master data, transactions, physical events and exceptions that define the current state.

02 / Reason

Understand the decision.

Supply the rules, history, constraints and evidence needed to produce a reviewable recommendation.

03 / Act

Return to the flow.

Route the approved decision into the system, portal, message or operator queue that can move the work.

Design principleThe shorter the distance between signal and controlled action, the more valuable the system becomes.

04Engagement

Start with one workflow. Scale what works.

An engagement can start at any stage, and every stage ends with something your team can use.

  1. 01

    Diagnostic assessment

    We go on site to map the workflow, data and decisions, and size the problems worth solving.

    You getA prioritized problem map with measurable targets

  2. 02

    Prototype sprint

    We build a working prototype with AI-assisted development and test it on your real data.

    You getA usable prototype, validated against those targets

  3. 03

    Embedded delivery

    Our engineers join your team and wire the solution into the live process, with human approval and rollback.

    You getA solution running in your operation

  4. 04

    Productize and operate

    We turn what repeats into a standard product, then keep improving and supporting it.

    You getA maintained product with regular iterations

  5. 05

    Training and enablement

    We coach your team to use AI tools well and to own the workflow.

    You getA team that runs and extends the system itself

05Field Proof

Delivered with our customers

Every product starts in a customer's live operation and is validated there before it becomes a standard capability. Our deepest engagement is with a large integrated logistics group; in manufacturing, we connect a textile producer's work from order to shipment.

First Exclusive Partner

A large integrated logistics group

A leading integrated logistics service group in South China

Source: public information on the customer's website
~1,000logistics professionals
30+branches
100,000+ m²warehouse space
RMB 5.5B+annual revenue
Live case work

One method. Two operating truths.

Explore how the same delivery discipline becomes a different operating system in logistics and manufacturing.

Exhibit 02Same delivery discipline. Different unit of truth.
Logistics & supply chainManufacturing
Primary unit of truthThe supply chainThe production order and batch
Work crossesCustomers, forwarders, carriers, customs, warehouses and driversPlanning, materials, production stages, quality and shipping
Critical evidenceRates, bookings, documents, milestones, handoffs and physical eventsSpecifications, work orders, lots, stage status, tests and deviations
Decision windowQuote, route, book, declare, move and recoverPlan, sequence, inspect, explain and correct
Failure to avoidA fast answer built on stale or fragmented supply-chain dataA recommendation detached from capacity, process or quality context
06Contact

Book a consultation, bring one real workflow, or discuss a partnership — email us.

Email usmgmt@huzhi.techSan Francisco · Irvine · New York · Hong Kong