$16.6 million financing challenges the trillion-dollar freight market: How Vooma turns AI Agents into digital employees for logistics companies
Jesse Buckingham
Vooma team
Original Statement
1. Track Background and Pain Points: The Digital Opportunity of the $900 Billion "Sleeping Giant"
1. A Huge and Critical Underlying Economic Pillar
• Market Size: The U.S. freight trucking industry is a massive $900 billion industry, expected to further rise to $1.46 trillion by 2035.
• The Operating Base of the Physical World: Almost every physical item consumed must go through hundreds of transfer touchpoints between shippers, brokers, and carrier truck fleets before reaching the end user.
2. Fundamental Reasons Why Traditional Software Cannot Automate
• Extreme Fragmentation and Chaotic Processes: The freight industry has long relied heavily on manual phone calls, emails, and filling out Excel spreadsheets.
• Limitations of Deterministic Code: Traditional software must be written based on strict, deterministic rules, while the processes of different companies in freight vary widely and unexpected situations frequently arise, leading to traditional automation tools being unable to achieve economic scalability due to high development and maintenance costs.
• Breakthrough of Large Language Models (LLM): Large models have the ability to flexibly handle complex problems in chaotic, unstructured text and voice environments, breaking the decades-long deadlock against automation in the freight industry.
2. Vooma's Core Product and Business Loop: "From Quote to Cash"
1. Financing and Capital Endorsement
• Background Journey: Graduated from YC (Y Combinator), focusing on intelligent architecture in the freight and logistics field.
• Capital Recognition: Completed over $16 million in financing, with the A round led by the well-known venture capital firm Craft Ventures, and the product has been deeply integrated and adopted by several large logistics companies in the U.S.
2. Full-Link AI Digital Employees (AI Co-workers)
The Vooma platform can build autonomous intelligent agents with digital identities that automatically execute full-process tasks across channels via email, phone voice, and SMS:
• Intelligent Quoting: Automatically captures and evaluates inquiries, quickly responds to shipper quotes, and assists freight forwarding companies in securing orders.
• Order Entry: Automatically transcribes unstructured cargo orders into the underlying Transportation Management System (TMS).
• Load Coverage: Automatically assigns and matches suitable carriers and contracted fleets.
• Scheduling: Coordinates with factories and warehouses to arrange truck arrival times for loading and unloading.
• Autonomous Tracking: Real-time access to truck GPS signals for dynamic monitoring of vehicle deviations, mid-route stops, or expected delays, autonomously triggering communication at critical nodes.
• Cash Collection: After unloading is completed, AI automatically calls to verify and urge drivers to collect proof of delivery documents, streamlining the payment process.
3. On-Site Phone Call Recording
• In the demonstration, an AI tracking agent named "Ashley" automatically called the truck driver who had just arrived in Atlantic City, New Jersey, to unload.
• The AI exhibited realistic natural voice interaction and adaptability, accurately inquiring and confirming the driver's actual arrival time (2 PM) and unloading departure time (5 PM), with the entire follow-up verification process efficiently completed in seconds, and the data directly stored.
3. R&D Rhythm and Customer Feedback: Rapid Iteration and Avoiding Technical Debt
1. Platform General Data Model Reconstruction
• Strategic Prioritization: Before customizing the automatic allocation of carrier contracts for specific large clients (such as VP Logistics), prioritize refining the underlying general data model to avoid getting trapped in specific client logic fixes too early.
• Aligning Abstract Cognition: In the definition of core entities, elevate the coverage strategy to a general top-level abstraction to support smooth migration for subsequent large-scale clients.
2. "Pulling Learnings Forward" Strategy
• Avoiding Go Live Traps: The past approach often pushed for immediate client deployment after product functionality was validated, leading to core missing functionalities being exposed months after system operation.
• Proactive Follow-Up on the Queue List: For the 10 companies waiting to onboard, proactively initiate in-depth communication to quickly identify functional gaps during the phase of a small user base and light historical backward compatibility burden, completing feature development in advance.
3. Team Culture Symbols
• "Chef" Ceremony on Mondays and Fridays: Every Friday, the most outstanding contributor of the week is selected as the "Chef," and the winner wears an apron and chef's hat, leading the team in a signature high-five during the Monday all-hands meeting to enhance cohesion.
• Viral AI Marketing: Using AI to generate a fun video of founder Jesse mimicking Jean-Claude Van Damme's "split" between two trucks, and inviting Jean-Claude Van Damme himself to record a humorous interaction on Cameo, achieving low-cost viral outreach.
4. Recruitment Assessment Reconstruction in the AI Era: Shifting from "Writing Code" to "Reviewing Code"
1. The Dilemma of Traditional Technical Interviews Failing
• Surge in Code Generation Proportion: In teams that heavily use AI tools, AI can now complete 80% of code writing. If interviews only assess whether candidates can handwrite basic code, they miss 80% of the true talent capability signals.
• Reevaluating Engineering Value: Companies no longer simply need junior coders who mechanically write logic but require composite engineers who can deeply harness AI and possess strong system architecture and discernment skills.
2. A New Engineering Assessment Paradigm
• Retain Basic Algorithm Testing: As a baseline to assess engineers' fundamental logical literacy.
• Introduce "PR Review Practice (Code Review / AI PRs)":
• Simulated Environment Setup: Establish a real monorepo containing two major projects: TypeScript (GraphQL architecture) and Python (traditional MVC REST API architecture).
• AI Generates Defective Code Merge Requests (PR): AI deliberately generates PRs containing architectural risks, logical flaws, or boundary condition defects.
• Assessment Focus: Evaluate whether candidates can efficiently identify architectural vulnerabilities, code smells, and propose optimizations through keen engineering intuition and collaboration with AI.
5. The Founder’s Technical Background and Early Entrepreneurial Experience
1. The Founder’s Cognitive Journey
• Background in Autonomous Driving Engineering: The founder has deep experience in the fields of autonomous passenger cars and trucks, which, despite being technically hardcore, have extremely long hardware iteration cycles.
• Transition to Software Intelligent Agents: Shifted to AI Agent architecture at the software level, capable of driving efficiency in trillion-dollar traditional industries at an agile pace, meeting enterprise-level clients' strict delivery requirements for "99.999% ultra-high accuracy."
2. Two Key Principles for Early Entrepreneurs
• Reduce Blind Following of Others' Advice: Early in entrepreneurship, it is easy to place hope in seeking ready-made answers from others, but truly excellent founders must possess the drive for independent thinking, personally understand the intricacies, and quickly respond to market feedback.
• Rapid Pivoting Until Locking in a Lifelong Mission: Early on, be bold in embracing great ambitions and be prepared to quickly overturn and iterate ideas based on market signals; once hitting upon a significant proposition worthy of dedicating a large part of one's life to, commit fully to becoming the top team globally in that field.
ABAB AI Insight
This material has a very strong direction, but there are three areas that must be corrected first, otherwise the title will "raise the company too high":
First, Vooma cannot currently be called a "unicorn". Public information does not show that it has reached a valuation of $1 billion; the official disclosure is a total financing of $16.6 million; the market database does not have a consistent and reliable public confirmation of the valuation.
Second, over $16 million is not the amount for Series A. Vooma has a total financing of $16.6 million, which consists of $3.6 million Seed + $13 million Series A; the Series A was led by Craft Ventures, and the Seed was led by Index Ventures.
Third, the so-called "trillion" refers to the future market size of the U.S. trucking industry, not Vooma's valuation. The U.S. trucking industry is expected to generate about $906 billion in revenue in 2024, and ATA and S&P Global predict it will reach about $1.46 trillion by 2035.
So I suggest a complete upgrade of the title.
My most recommended title:
$16.6 million financing challenges the trillion-dollar freight market: How Vooma turns AI Agents into digital employees for logistics companies.
A trend-based title could be:
From phone calls and emails to AI Co-workers: How Vooma rewrites the $1.46 trillion U.S. freight workflow.
An entrepreneurial title could be:
Self-driving engineer × logistics software CEO: How Vooma uses AI Agents to penetrate the hardest-to-digitize traditional industry.
This is more accurate than "AI Agent unicorn" and actually more powerful.
────────────────
1. What Vooma is truly worth studying is not "AI helping logistics companies save a few people."
The real big logic is:
For the past 30 years, software has mainly been responsible for recording what humans have done; in the Agent era, software begins to directly complete work for humans.
This is a shift from:
System of Record
to:
System of Action.
These two business models are not on the same scale.
In the past, TMS told you:
Where the cargo is,
How much it costs,
Who the driver is.
In the future, what Vooma wants to do is:
Send emails by itself,
Make phone calls by itself,
Quote by itself,
Find trucks by itself,
Schedule by itself,
Track by itself,
Upgrade exceptions by itself.
This is no longer just pure SaaS.
Instead, it is selling:
Digital Labor.
────────────────
2. Why is freight one of the most attractive landing industries for AI Agents?
Because it meets five conditions at the same time:
First:
The market is huge.
2024 U.S. trucking revenue:
About $906 billion.
Predicted by 2035:
About $1.46 trillion.
Second:
Labor-intensive.
Third:
Highly repetitive processes.
Fourth:
Extremely unstructured inputs.
Fifth:
Automation ROI can be directly calculated in money.
These five conditions together are very rare.
────────────────
3. Moreover, the U.S. freight industry has a particularly critical structure: extreme fragmentation.
In 2024, among U.S. carriers:
91.5% have 10 or fewer trucks.
99.3% have fewer than 100 power units.
This means the entire industry is not:
5 super companies dominating the entire market.
But rather:
A large number of small carriers,
A large number of brokers,
A large number of shippers,
A large number of warehouses,
A large number of different systems
pieced together.
Fragmentation means:
Coordination Costs are extremely high.
And what AI Agents prefer is precisely:
High coordination cost industries.
────────────────
4. What is truly expensive in logistics is often not the trucks, but the "coordination."
A load from the moment the customer raises a demand may go through:
Quoting,
Confirmation,
Order entry,
Finding carriers,
Checking qualifications,
Negotiating prices,
Scheduling pickups,
Scheduling deliveries,
Driver check-calls,
GPS tracking,
Exception handling,
POD,
Billing.
Each step may require:
Email.
Phone.
Text.
Excel.
TMS.
Portal.
Thus, a large amount of manpower is not spent on:
"Creating logistics."
But rather on:
Moving information.
────────────────
5. This is the core investment logic of Vooma.
Trucks will continue to be driven by people.
Warehouses will continue to exist.
Goods will continue to need to be physically moved.
However:
The information flow around the goods can be largely consumed by AI.
This is a very big difference.
In the past, logistics technology mainly optimized:
Physical Movement.
Now AI begins to optimize:
Information Movement.
And many times:
If the information is slow,
It will ultimately lead to the goods being slow as well.
────────────────
6. This is the manifestation of the Theory of Constraints in logistics.
For example:
A customer email comes in.
It takes 20 minutes for a person to quote.
20 minutes late for confirmation.
20 minutes late to build the load.
20 minutes late to post the load.
The good truck has already been taken by someone else.
In the end, you can only use a more expensive carrier.
So a problem that seems to be:
"Just administrative efficiency"
ultimately affects:
Gross Margin.
So AI is not just about:
Saving 5 minutes.
It may change:
Revenue + Margin + Service Level.
────────────────
7. This is also why the value of Vooma's Quote product is much greater than "automatically writing emails."
If AI can make quoting:
From a few minutes,
To a few seconds,
Theoretically it could increase:
Quote Volume.
At the same time improve:
Response Speed.
Vooma's current self-reported customer results include:
About 15% email quote win-rate improvement,
Saving more than 2 hours per rep per day,
Data errors reduced by over 90%, etc.
These belong to the product effects disclosed by Vooma itself and should not be considered independent audit data, but they indicate that it is selling products based on business metrics rather than "AI is cool."
────────────────
8. Here we see the biggest pricing change between AI SaaS and traditional SaaS.
Traditional SaaS often sells:
Seat.
100 employees.
Each:
$50/month.
Company revenue:
$5,000/month.
But if AI Agents really complete:
The work of 10 employees,
Customers will not only measure value based on:
"Software seat"
They will start measuring based on:
Labor Replacement / Revenue Outcome.
────────────────
9. This is also why the TAM of Agentic Software may be much larger than traditional SaaS TAM.
Traditional software competes for:
IT Budget.
AI Agents begin to compete for:
Labor Budget.
This is a completely different market.
For example, a certain process originally:
20 people × $70K total cost
=
$1.4M / year.
If AI can complete 50% of the work,
The theoretical value pool is not:
$50/seat/month.
But hundreds of thousands or even millions of dollars.
────────────────
10. This is the most important statement about future AI enterprise software.
Software is eating labor spend.
Marc Andreessen once said:
Software is eating the world.
A more precise statement for the Agent era might be:
AI software is eating workflows, and workflows are mostly labor costs.
────────────────
11. Vooma's initial entry into Order Entry is a very smart wedge.
When YC first introduced Vooma, its product was far from what it is today.
The initial core problem was:
Order Entry.
Customers sent loads through:
Emails,
PDFs,
Excel,
Screenshots.
Employees manually read them out,
And filled them into TMS.
Vooma:
Reads,
Extracts,
Structures,
Writes into the system through EDI / API.
This is a very typical:
Wedge Product Strategy.
────────────────
12. Why not say from the beginning "I want to reconstruct global logistics"?
Because customers will not buy it.
One of the biggest mistakes entrepreneurs make is:
Vision is too big,
Product has no entry point.
Vooma is very smart:
Vision:
AI Agent for Logistics.
The first step:
Help you reduce a few order entries.
Painful enough.
Simple enough.
ROI can be calculated.
Easy to implement.
────────────────
13. Once entering the workflow, then expand sideways.
Order Entry
↓
Quote
↓
Schedule
↓
Cover
↓
Track
↓
POD / downstream workflows.
Now Vooma's official website has directly defined the product as:
A complete platform from quote to cash.
This is the classic:
Land and Expand.
────────────────
14. So its most important strategy is not to "make many Agents"
But to:
Occupy the entire Workflow Graph.
If you only do:
Quote,
Competitors can replace you.
If you only do:
Voice,
You may also be replaced.
But if:
The data generated from Quote
Enters Build,
Build enters Schedule,
Schedule enters Cover,
Cover enters Track,
Track then affects payment,
forming continuity of state between systems,
and the switching costs begin to rise.
────────────────
15. The truly powerful AI companies will ultimately compete for "system state"
This is much more important than chat capabilities.
For example, an Agent knows:
Who the customer is,
What this shipment is,
What the past quotes were,
Who the carrier is,
When the appointment is,
Whether the driver is late,
Whether payment is completed.
This Agent can truly:
Act.
If every time it starts from scratch asking:
"Please tell me the context again."
Then it is still just a chatbot.
────────────────
16. Therefore, the true moat of an Agent is not LLM
OpenAI,
Anthropic,
Google
models will only get stronger in the future.
Vooma will find it hard to rely on:
"Our large model is better than others"
to establish a permanent moat.
The real moat should come from:
Workflow Context.
Integrations.
Domain Ontology.
Customer SOPs.
Historical Outcomes.
Actions.
These six things.
────────────────
17. Especially the Tribal Knowledge in logistics
This is a term Vooma emphasizes repeatedly.
Many logistics company rules are not written into software.
They exist in the minds of employees.
For example:
This customer cannot pick up after 4 PM on Fridays.
This warehouse is always 30 minutes late.
This carrier is cheap but should not be given high-value goods.
This customer adds a certain margin when quoting.
This lane needs to pay attention to the weather in certain seasons.
These are:
Tribal Knowledge.
────────────────
18. Why has this kind of knowledge been hard to software?
Because traditional programs require:
Explicit Rules.
IF A,
THEN B.
But real business may have:
Thousands of small exceptions.
Different customers are different.
Different facilities are different.
Different seasons are different.
Traditional engineers end up writing:
if statements repeatedly.
────────────────
19. So what does software end up becoming?
Exception Spaghetti.
A lot of:
Customer A special rules,
Customer B patch,
Customer C workaround.
Every time a new customer is added:
Complexity increases.
In the end, maintenance costs explode.
This is the fate of many traditional enterprise automation.
────────────────
20. LLM has truly changed this cost curve for the first time
Because LLM can understand:
Natural language,
Emails,
PDFs,
Phone calls,
Non-standard expressions.
So many tasks that were previously:
"Theoretically automatable, but not cost-effective in engineering"
suddenly become:
Economically Automatable.
This term is particularly important.
It’s not that previous technology was absolutely incapable.
But rather:
Automation cost > Labor cost.
LLM has flipped this line.
────────────────
21. This is the essence of the technological revolution Vooma is in
It’s not:
AI is smarter.
But rather:
Software Coverage Expansion.
Previously, software could only cover:
50% of clearly defined rules.
The remaining:
50% handled by people.
Now AI may allow software to cover:
70%,
80%,
90%.
Thus, the entire enterprise work design needs to be restructured.
────────────────
22. This is why Mike Carter's background in autonomous driving is actually very suitable for this company
Mike was a founding engineer at Kodiak Robotics and led the motion planning and safety team; previously, he worked on autonomous driving at Uber ATG and Otto.
This sounds like:
Jumping from autonomous driving to SaaS.
In fact, the underlying problems are highly similar.
────────────────
23. What does autonomous driving really solve?
The real world:
Is very chaotic.
There are:
Noise,
Anomalies,
Uncertainty,
Safety requirements.
The system must:
Perceive,
Judge,
Act.
Logistics Agents are the same:
Emails are non-standard.
Phone conversations differ.
Carriers change their minds.
Drivers are delayed.
Warehouses do not respond.
Quotes change.
The system must also:
Perceive → Reason → Act.
────────────────
24. So Mike's transition from Autonomous Vehicle to Autonomous Workflow is actually a continuous path
The first type of autonomy:
Allows machines to act in the physical world.
The second type:
Allows machines to act in enterprise workflows.
The former needs to control:
The steering wheel.
The latter needs to control:
Email,
Phone,
TMS,
Portal.
The underlying common problem is:
Reliability under uncertainty.
────────────────
25. But your "founder has deep experience in autonomous driving" needs to be slightly modified
Vooma has a dual-founder structure.
Mike Carter is a veteran in autonomous driving technology.
Jesse Buckingham has a completely different background:
Bain consultant,
Stanford MBA,
Later served as CEO of PE-backed logistics software company ASG LogisTech, growing the business from about $2 million ARR to over $20 million.
This is actually more impressive than two pure AI engineers starting a company.
────────────────
26. Because Vooma's founding team has a very classic combination
Mike:
Technology Truth.
Understands automation,
AI,
Safety,
Systems.
Jesse:
Market Truth.
Understands logistics customers,
Sales,
Software deployment,
Economic models.
This is called:
Founder-Market Fit.
────────────────
27. Many Vertical AI Startups fail because they only have one half
Technical teams:
Strong in AI.
But do not understand the real workflows of customers.
In the end, they produce:
Beautiful demos,
But no one buys.
Traditional industry entrepreneurs:
Know the pain points.
But lack the ability to make the Agent reliable enough.
In the end:
They are still just consulting firms.
Vooma's greatest potential lies in:
This combination of two capabilities.
────────────────
28. And now there are already quite obvious early PMF signals
Vooma officially disclosed that in the previous year:
Revenue grew by about:
12.5x.
Transaction volume:
32x+.
And handled hundreds of thousands of loads.
Customers include:
Echo Global Logistics,
Arrive Logistics,
MODE Global,
NFI Logistics,
MoLo / ArcBest,
Sunset, etc.
These are still within the company's disclosures, and specific ARR has not been made public.
So:
There are signals of PMF, but the revenue scale cannot be publicly confirmed.
This must be distinguished.
────────────────
29. Therefore, calling it a "unicorn" now is completely too early
Vooma's current more accurate positioning is:
High-growth Series A Vertical AI Startup.
This is actually more worth observing.
Because truly potentially huge companies,
The most valuable research time point is usually not:
Already $10B.
But rather:
Still only a few dozen people.
YC's current page shows a team of about 40 people.
────────────────
30. What Vooma really needs to prove now is moving from "Feature PMF" to "Platform PMF"
In the early stage:
Customers are willing to use Build.
This is:
Feature PMF.
Later:
Customers use Quote, Cover, Schedule, Track simultaneously.
Starting to form:
Platform PMF.
This transition is extremely important.
Because Features can easily be:
TMS,
Other AI Startups,
Microsoft,
Model companies
copied.
────────────────
31. What is truly difficult to replicate about a platform is "business memory"
Vooma now emphasizes:
The system learns each enterprise's:
SOP,
Customers,
Carriers,
Facilities,
Historical experiences.
If it really accumulates,
What it ultimately builds is not:
An Agent.
But rather:
Company Operating Memory.
This concept is very significant.
────────────────
32. One of the largest implicit assets of enterprises in the past has always existed in the minds of employees
When employees leave:
Knowledge goes with them.
Newcomers:
Learn again.
As enterprises expand:
Experience transfer becomes slower and slower.
If AI can consolidate:
The SOPs,
Judgments,
Exception handling methods
of excellent employees,
For the first time, it may form:
Institutional Intelligence.
────────────────
33. This is also one of the reasons I am optimistic about Vertical AI
Horizontal AI knows:
General knowledge worldwide.
Vertical AI should really know:
How this company actually works.
These two types of Intelligence are completely different.
The former:
General Intelligence.
The latter:
Operational Intelligence.
What really allows companies to spend big money,
is often the second type.
────────────────
34. Vooma's Cover product has now crossed an interesting boundary
It is not just:
answering calls.
The current product can:
screen carriers,
verify qualifications,
answer questions,
take inbound calls 24/7,
and even perform:
automated rate negotiation.
This begins to enter the realm of true:
Economic Decision.
────────────────
35. Why is this much harder than "help me summarize an email"?
If AI mis-summarizes an email:
a human can just look at it.
If AI misreports a freight rate:
it could directly lose margin.
If AI selects the wrong carrier:
it could cause:
accidents,
theft,
delays,
insurance risks.
So as the Vertical Agent goes deeper into the value chain:
Error Cost increases.
────────────────
36. This is one of the most critical formulas in the Agent industry
AI Value = Task Value × Automation Rate − Error Cost
If the task value is very low:
it is not worth doing.
If the error cost is extremely high:
it may not be fully automated for now.
The truly beautiful scenario is:
high frequency,
high labor costs,
verifiable results,
and layered risks.
Many processes in freight meet this criterion.
────────────────
37. Therefore, a true enterprise-level Agent should not pursue "100% autonomy"
but should design:
Autonomy Gradient.
Simple cases:
AI completes automatically.
Medium cases:
AI completes + human confirmation.
High-risk cases:
Escalate to Human.
This is much more mature than just saying:
"fully unmanned."
────────────────
38. Vooma Track has clearly adopted this design
The official product process is:
AI contacts drivers based on rules,
through:
email,
text,
voice
to obtain status,
and writes back to TMS;
if it reaches the abnormal conditions defined by the customer,
it escalates to:
dispatcher or internal team.
This is the correct:
Human-in-the-Loop Architecture.
────────────────
39. This also explains why the first principle of enterprise Agents is not Intelligence, but Reliability
Consumer chat:
95% accuracy,
is already impressive.
Financial transfers:
95% accuracy,
is a disaster.
Healthcare:
95% accuracy.
Disaster.
Key processes in enterprise logistics:
are the same.
So what a true B2B Agent needs to optimize is not:
"how smart it looks."
But rather:
"is it controllable when it fails."
────────────────
40. What enterprises really need is actually five layers of reliability
First:
Detection
Knowing when it is uncertain.
Second:
Guardrails
Cannot exceed permissions.
Third:
Escalation
Knowing when to seek human help.
Fourth:
Audit Trail
Being able to check what happened later.
Fifth:
Observability
Managers knowing what the Agent did today.
If these five layers are not established,
AI Co-workers will find it hard to truly enter the core systems of Fortune 500.
────────────────
41. Therefore, descriptions like "99.999% accuracy" should not be directly taken as Vooma's formal SLA
From the public introduction of INSIDE Startups, it can be confirmed that:
large logistics clients have extremely high requirements for software accuracy and reliability, and the cost of errors is very high.
But I did not find Vooma's official public commitment:
99.999% accuracy
as this SLA.
So it is recommended to change it to:
Enterprise clients require reliability close to critical task levels.
This is much more prudent.
────────────────
42. The data model reconstruction you mentioned is actually one of the most easily overlooked aspects of AI entrepreneurship
Outsiders think making an Agent is:
Write the prompt well.
Connect the API.
Done.
Not at all.
One of the hardest things in Enterprise AI is:
Ontology.
That is:
how the system defines the world.
────────────────
43. For example, what exactly is a Load in logistics?
A Load includes:
shipper,
consignee,
pickup,
delivery,
carrier,
rate,
appointment,
status,
documents.
Then there are:
coverage strategy,
exceptions,
customer SOP.
If the bottom-level object design is wrong,
as the Agent does more and more,
the entire system will become increasingly chaotic.
────────────────
44. This is why data models are closer to a moat than UI
UI:
can be redone in a month.
Data Model:
once hundreds of clients rely on it,
it is very hard to change.
Because it will affect:
API,
historical data,
automation,
workflow,
reporting,
integrations.
This is what is called:
Schema Lock-In.
────────────────
45. "Pull Learnings Forward" is also a very good entrepreneurial idea
Most startup teams:
Customer 1 goes live.
Problems arise.
Fix it.
Customer 2 goes live.
More problems arise.
Fix it.
Constantly:
Reactive.
A smarter approach is:
Before a large-scale go-live,
bring in queued customers to study:
What are we definitely going to run into in the future?
This is called:
Pre-mortem Product Development.
────────────────
46. This is especially important in Enterprise Software
Consumer Apps:
If there is a bug,
issue a patch.
Enterprise systems:
After going live:
data migration,
training,
SOP,
permissions,
integration
are all established.
Changing core abstractions afterward,
is very costly.
So early on, it is essential to leverage:
Low Backward-Compatibility Cost Window.
────────────────
47. In the early stages of entrepreneurship, there is a very valuable asset that many people do not realize
is:
You are allowed to break things.
Few customers.
Few APIs.
Little historical data.
Few dependencies.
So you can:
change the database,
change the product,
overhaul the architecture.
Once the company grows,
every design flaw will turn into:
Legacy Debt.
────────────────
48. Therefore, excellent entrepreneurs in the early stages should not only pursue Speed
but should pursue:
Speed of Learning.
This is not entirely the same as:
Speed of Shipping.
Doing 10 features,
without learning anything,
has no value.
Doing 1 feature,
discovering that the underlying model is wrong,
and then restructuring in advance,
can have enormous value.
────────────────
49. Changes in recruitment methods are also worth high attention
Mike Carter has recently publicly confirmed:
Vooma now requires engineering candidates to undergo:
Code Review Interview,
focusing on testing candidates' understanding and judgment of:
AI-generated decisions,
AI code.
I think this direction is very important.
────────────────
50. Because "writing code" is rapidly transitioning from a scarce skill to a relatively cheap skill
In the past:
Engineer Value
≈
how much correct code they can write.
In the future, it will increasingly approach:
Engineer Value
≈
how to define what should be built + judging whether AI-written things are garbage.
This has caused a huge shift in value.
────────────────
51. The most valuable skill for engineers in the future may shift from Coding to Judgment
Including:
architectural judgment,
security judgment,
boundary conditions,
performance,
abstraction levels,
data models,
technical debt,
product trade-offs.
Because AI is very good at:
Generate.
However:
choosing which generation is correct,
still requires:
Taste.
────────────────
52. This is very similar to the changes happening in the investment industry
In the past, an analyst's value was:
finding data,
doing Excel,
building models.
AI is increasingly capable of doing all of this.
In the future, what will truly be valuable is:
Which assumptions matter?
Similarly:
programmers in the future will not compete on:
how fast they can type.
But rather:
who knows which code should not be written at all.
────────────────
53. Therefore, the so-called "AI writes 80% of the code" should not be mechanically understood as "80% of engineers will be unemployed"
Because code production is not all of software engineering.
AI reduces the:
Cost of Code.
But it may lead to a significant increase in the:
Amount of Software.
This is similar to:
Jevons Paradox.
The cheaper programming becomes,
the more likely companies will need more software.
────────────────
54. What may truly be compressed is a specific type of work
Pure Implementation Labor.
That is:
Others have already defined the problem very clearly,
You are just translating the spec into code.
This layer is the most dangerous.
What will be most valuable in the future is:
Problem Definition,
Architecture,
Review,
Integration,
Ownership.
────────────────
55. So what should really be tested in hiring during the AI era are three things:
First:
Can you discover AI's mistakes?
Second:
Can you judge what is worth doing?
Third:
After AI makes a mistake, can you take ultimate responsibility?
This is:
Accountability.
AI has no career.
Engineers do.
────────────────
56. The "Chef" culture seems funny, but it is actually very typical.
Vooma indeed has internally:
Chef
Awards, publicly posting employee mentions to reward the team or member who "cooked hardest this week."
This cultural symbol is very effective for a startup of about 40 people.
Because early culture cannot just be written on the wall:
Integrity,
Innovation,
Customer First.
No one remembers.
────────────────
57. A truly strong company culture must be able to turn into behavioral language.
For example:
"Did you cook this week?"
The team immediately knows:
It means:
Shipping,
Ownership,
Impact.
Netflix has:
Keeper Test.
Amazon has:
Day 1.
Meta has:
Move Fast.
The most powerful state of culture is not:
Values written down.
But:
Values compressed into language.
────────────────
58. But the real danger of cultural rituals lies here.
If the rewards are:
Longest working hours,
Who sleeps the least,
Who is the craziest,
It can easily turn into:
Performative Hustle.
What should really be rewarded is:
Output / Learning / Customer Impact.
Not:
Pain.
This is actually the same logic as Doug Leone's "suffering" mentioned earlier.
Do not mistake:
Suffering
for:
Productivity.
────────────────
59. The marketing case of Jean-Claude Van Damme is also very clever.
Vooma indeed remade the famous "double truck split" ad of Volvo Trucks into an AI/mixed version, using Jesse as the protagonist to promote Cover Outbound. The relevant production team publicly introduced this content experiment.
What is truly worth learning is not:
"Funny."
But:
Category-Native Marketing.
────────────────
60. Why is this idea particularly suitable for Vooma?
Because the original ad itself is:
Classic content of truck culture.
Target customers:
freight brokers,
carriers,
logistics people.
They understand the joke at a glance.
This is 100 times stronger than an ordinary AI startup shooting:
"A blue robot flying in a data center."
────────────────
61. A huge misconception in B2B branding is that "it must be serious."
In fact, B2B buyers are also human.
Especially Vertical SaaS:
The industry circle is very small.
If your content can become:
Industry Meme,
the customer acquisition efficiency may be very high.
This is called:
Earned Distribution.
It is very important for startups with only a few tens of millions of dollars in funding.
────────────────
62. Now back to the company's real business ceiling: Will Vooma become the "Salesforce of logistics"?
I think this metaphor is not entirely accurate.
The core of Salesforce:
Human Workflow Management.
Vooma is more likely to move towards:
Autonomous Workflow Execution.
So it is closer to:
ServiceNow + AI Worker + Vertical Operating Layer.
────────────────
63. Vooma's real biggest opportunity is to become the Freight Control Plane.
In the future, brokers will open an interface,
and what they see will no longer be:
50 employees handling 50 things themselves.
But:
20 employees
200 Agents.
Employee management:
Exceptions,
Relationships,
Strategy.
Agent management:
Routine Execution.
This is what the founder describes as:
swarms of agents.
────────────────
64. This will change the organizational structure of a logistics company.
In the past, business scale was roughly:
Revenue ∝ Headcount.
More cargo,
means hiring more people.
If the Agent succeeds:
Revenue can grow,
Headcount grows more slowly.
Thus:
Revenue per Employee
increases.
This is the metric that CEOs are truly willing to pay for.
────────────────
65. So the KPI that Vooma should ultimately sell is not "time savings"
but:
Loads per Employee.
Revenue per Rep.
Gross Margin per Load.
Quote Win Rate.
Cost per Load.
Exception Rate.
If these continue to improve,
Vooma will be very valuable.
────────────────
66. This is the biggest change in sales philosophy between AI companies and traditional SaaS.
Traditional SaaS:
I have 100 features.
AI Agent:
I help you move 20% more freight without hiring 20% more people.
The latter is:
Outcome Selling.
Not:
Feature Selling.
────────────────
67. But Vooma's biggest risk also comes from here: the value is huge, so competition must be fierce.
This track will definitely not only have Vooma.
Future competition will come from:
Vertical AI startups,
TMS incumbents,
freight-tech platforms,
voice-agent companies,
and even:
large logistics companies developing AI themselves.
So:
"A large market"
is never a moat.
The larger the market:
the more competition there usually is.
────────────────
68. The first moat it really must establish: Integration Density.
Now Vooma has already connected:
McLeod,
Turvo,
DAT,
Truckstop,
Highway,
Front,
RingCentral,
Slack
and many other industry systems.
With each additional integration,
the product:
becomes easier to deploy.
With each additional customer:
integration ROI increases again.
This will form a certain:
Ecosystem Moat.
────────────────
69. The second moat: Proprietary Workflow Data.
For example:
What prices are easy to close?
Which carriers are more reliable on which lanes?
What time slots are more likely to succeed in booking?
What emails usually indicate exceptions?
What communication methods are more likely to get responses?
If this data can enhance model capabilities while protecting customer privacy and permissions,
the long-term value is very large.
────────────────
70. The third moat: Trust.
This is something that enterprise AI is most easily underestimated.
Logistics CEOs need to ask:
Do I dare to let this Agent:
Directly quote to customers?
Directly negotiate with carriers?
Directly update TMS?
Directly call drivers?
The real competition is not just:
A 2% higher model score.
But:
Who do I trust with execution?
────────────────
71. Therefore, Enterprise AI will exhibit a very strong Winner-Take-Most dynamic.
Once a company has allowed Vooma to:
Read all emails,
Connect to TMS,
Understand SOP,
Know customers,
Know carriers,
Master workflows,
The cost of replacing it with another AI startup begins to be very high.
This is called:
Operational Switching Cost.
────────────────
72. The fourth moat: Distribution.
The biggest difference between Vertical AI and ordinary internet startups is:
Limited customers.
Small circle.
Reputation is extremely important.
If:
Echo,
NFI,
Arrive,
MODE
and other industry companies use it,
other brokers will take notice.
This is called:
Referenceability.
It is extremely valuable in B2B Enterprise sales.
────────────────
73. The fifth moat may actually be "people."
Vooma is still in the stage of a few dozen people.
The biggest risk at this stage is usually not:
Google snatching the market.
But rather:
Can we quickly hire the right people?
Especially:
Forward-Deployed Engineer,
Deployment,
Customer Success
will be extremely important.
Because every enterprise process is different,
Agents must be deployed.
────────────────
74. This is why AI Agent startups are reigniting the Forward-Deployed Engineer.
Palantir has long proven that:
Enterprise software truly needs to be deployed,
and cannot just throw APIs at customers.
You need engineers:
to go in,
understand the business,
connect data,
modify workflows,
and solve exceptions.
In the AI era, this position is even more valuable.
────────────────
75. So, could "AI companies" ultimately resemble consulting firms more than traditional SaaS?
In the early days:
A little bit.
Because it is necessary:
to deeply deploy.
But the key question is:
Can customization be productized?
If every new customer requires:
20 engineers for six months,
it’s just tech consulting.
If every deployment experience can:
be distilled into reusable primitives,
the company can scale.
────────────────
76. This is where your material's "first build a general data model" becomes truly important.
Every customer customization must ask:
Is this:
A one-off request?
Or:
A generalizable capability?
Excellent product teams continuously convert:
customer exceptions
into:
platform primitives.
This is the process of enterprise software transitioning from project to platform.
────────────────
77. From a VC perspective, when researching Vooma, I wouldn’t first ask "Is the market $1.46 trillion?"
That’s no longer up for debate.
I would ask seven questions.
First:
What is the ARR?
Currently undisclosed.
Second:
What is the Net Revenue Retention rate?
Third:
How many agents does a customer use in a year?
Fourth:
What is the Gross Margin?
Especially will a large amount of voice/LLM inference compress margins?
Fifth:
What is the Implementation Cost?
Sixth:
Is the Human Oversight Ratio continuously decreasing?
Seventh:
Can the cross-sell from Build to full platform work?
These determine the valuation.
────────────────
78. Especially Gross Margin is a problem all Agent companies must face in the future.
Traditional SaaS:
Low server costs.
AI Agents:
Each task may generate:
LLM tokens,
voice inference,
telephony,
tool calls,
third-party APIs,
real-time computation.
Thus:
Revenue growth
does not equal:
Gross Profit growing at the same rate.
────────────────
79. Therefore, Agent companies must focus on a very important new metric:
Cost per Completed Task.
For example:
Completing a load coverage:
Charge the customer:
$10.
Model,
phone,
API costs:
$2.
Gross Profit:
$8.
Very good.
If AI continuously retries to complete a complex task,
costs:
$9.
That’s dangerous.
────────────────
80. In the future, the technical capabilities of Agent companies will directly translate into Financial Margin.
Smarter model routing:
cheaper.
More effective prompts:
cheaper.
Better caching:
cheaper.
Small models can do what large models do not need to:
cheaper.
Lower voice latency:
better.
So:
AI Infrastructure Optimization = Gross Margin Optimization.
Technology and finance begin to bind directly.
────────────────
81. There is also a huge risk: Incumbent Bundling.
Suppose a certain TMS launches:
AI Build,
AI Quote,
AI Track.
Then includes them for free in:
existing contracts.
What will Vooma do?
This is a problem any application-layer AI startup must face.
────────────────
82. The answer cannot just be "Our AI is better."
The real answer must be:
Cross-System Neutrality.
TMS only understands:
their own systems.
If Vooma can cross:
email,
voice,
DAT,
TMS,
Highway,
portal,
Slack
to become a neutral coordination layer,
it has the opportunity to stand:
above the systems.
────────────────
83. This actually resonates beautifully with our recent study of FedEx.
FedEx wants to become:
A Global Supply Chain Orchestrator.
Vooma wants to become:
A Freight Workflow Orchestrator.
The difference is:
FedEx controls:
The Physical Network.
Vooma controls:
The Digital Workflow.
One manages goods.
The other manages the human coordination work around goods.
────────────────
84. The truly huge opportunity in the AI era may lie between the two.
The real world has:
factories,
trucks,
warehouses,
airplanes,
ports.
The digital world has:
Agents.
One of the companies with the greatest future value will likely be responsible for:
mapping digital intelligence to physical world actions.
Logistics is the most natural testing ground.
────────────────
85. This is also why I dislike the simplistic statement "AI disrupts logistics."
Trucks will not disappear because of Vooma.
Drivers will not vanish overnight.
Brokers will not completely disappear either.
What really happens is:
The division of labor is redesigned.
AI handles:
routine coordination.
Humans handle:
relationships,
exceptions,
negotiation edge cases,
strategic capacity,
customer trust.
────────────────
86. In the future, the best Freight Rep may not handle 100 calls themselves
but rather:
manage a group of AI Workers.
They need to:
define rules,
judge exceptions,
optimize Agents,
check metrics,
manage relationships.
This is like after the Industrial Revolution:
The most valuable are no longer purely manual workers,
but:
those who manage machines.
────────────────
87. So what Vooma is really selling can be further abstracted to:
Management Leverage.
In the past, one manager:
managed 10 people.
In the future:
10 people + 100 Agents.
If output increases fivefold,
this is:
The true enterprise value of AI.
────────────────
88. From a startup perspective, Vooma has another particularly valuable lesson.
It did not look for problems from:
"The AI industry."
It looked for problems from:
"The logistics industry."
This is what many AI entrepreneurs should learn the most.
The wrong question:
What can AI do now?
The better question:
In which multi-billion dollar industry are there large amounts of expensive, repetitive, unstructured work?
Then ask:
Can AI finally solve it now?
────────────────
89. This is the correct entrepreneurial sequence for Vertical AI.
Industry Pain
↓
Workflow
↓
Economic Value
↓
AI Capability
↓
Product.
Not:
LLM
↓
Think of a Demo
↓
Find someone to buy.
The order is completely different.
────────────────
90. Moreover, Vooma's dual-founder structure itself is the most classic entrepreneurial lesson.
One person knows:
The Problem.
One person knows:
The Technology.
Many excellent Vertical AI companies will grow into this combination.
Healthcare:
Doctor + AI Engineer.
Law:
Lawyer + AI Engineer.
Logistics:
Operator + Autonomy Engineer.
Finance:
Banker + AI Engineer.
This is a very important entrepreneurial paradigm for the coming years.
────────────────
91. From the perspective of capital allocators, I would not currently regard Vooma as "a successful company."
It is far from it.
It is:
Series A.
Dozens of people.
Total funding:
$16.6M.
Valuation public information is limited.
It is far from:
IPO,
$1B valuation,
industry dominance.
So we cannot use results to infer:
"It will definitely succeed."
────────────────
92. But it is worth studying because it stands at the intersection of three huge structural changes.
First:
AI Agents.
Software begins to act.
Second:
Vertical AI.
Industry knowledge enters the model.
Third:
Industrial Digitization.
Traditional industries in the real world begin to be reconstructed by AI.
If these three forces successfully overlap,
it will create very large companies.
────────────────
93. If Vooma ultimately fails, where is it most likely to die?
I would focus on five things.
1. Reliability is insufficient.
The demo is great,
but production is not.
2. Deployment is too heavy.
Every customer has to redo the project.
3. Incumbents copy features.
TMS directly bundles.
4. AI unit economics are not established.
The costs of reasoning and voice are too high.
5. Agents are not deep enough.
In the end, it's just "better RPA."
These are the five core risks.
────────────────
94. If it succeeds, the truly successful form will never just be "logistics robots making calls."
What should ultimately emerge:
Persistent Agents.
Long-term existence.
Having identity.
Having authority.
Having memory.
Having KPIs.
Even able to:
Invoke each other.
For example:
Quote Agent
Notifies:
Cover Agent.
Cover Agent
Notifies:
Schedule Agent.
Schedule Agent
Notifies:
Track Agent.
This is the true:
Agentic Enterprise.
────────────────
95. At that time, organizational structures may even see new concepts.
Today the CEO asks:
How many employees?
In the future, they may ask:
Human Headcount
Agent Count.
Today the CFO calculates:
Payroll.
In the future, they may also calculate:
Digital Labor Spend.
This is one of the biggest changes in enterprise software in the next five to ten years.
────────────────
96. And this is why Vooma's story is more worth watching than an ordinary logistics SaaS.
It is not simply about increasing:
Software penetration.
It is testing a larger proposition:
Can the dozens of actions a knowledge worker completes in a day be autonomously executed by software?
If the answer is:
Yes,
The revenue pool for software companies will expand from:
IT Budget
to:
Global Labor Budget.
This is the true greatness of AI Agents.
────────────────
Finally, to compress the entire Vooma case into one sentence:
What Vooma is really betting on is not whether "AI can help logistics companies write emails," but whether the organizational structure of logistics companies will shift from "humans using software" to "humans managing AI Workers"; once the latter is established, the market for AI software will no longer just be the SaaS budget, but will begin to consume hundreds of billions in labor and coordination costs.
From an entrepreneurial perspective, I think the most valuable lesson is:
Do not look for a demo suitable for AI, but look for a huge industry that becomes "economically automatable" for the first time due to the emergence of AI.
The U.S. freight industry is such an industry:
A scale close to $1 trillion,
Expected to reach $1.46 trillion by 2035,
Still heavily reliant on phone calls, emails, Excel, and manual coordination for decades.
So the real question worth observing about Vooma is no longer:
"Can AI do logistics?"
But rather:
"Can Vooma become the AI labor operating system for the logistics industry?"
This will determine whether it will only be a multi-billion dollar Vertical SaaS or grow into a larger infrastructure company.
Vooma is still very early, and financing, product expansion, and major client contracts will change rapidly, so continuous tracking will be more valuable.
J