Dream Bigger: Using AI for Business Growth in Manufacturing

For twenty years the honest answer to most growth ideas in a fab shop was “great idea, who’s going to do it?” That answer is expiring. Today you can automate almost any office process, build almost any piece of software your shop needs, and stand up a small army of AI agents to chase the paperwork. The hard part is no longer can the tech do it. It’s what did we always want to do and never had the people for? Capability is getting cheap. Imagination is the new bottleneck. So: dream bigger.

This post is about AI for business growth, not cost cutting: four questions every owner should ask, the honest limits, and a five-question “dream list” to fill in this week.

Why is imagination now the bottleneck, not the technology?

Two curves explain it.

AI is getting cheaper faster than any technology in history. Epoch AI’s September 2026 report, The plunging price of thought, estimates the cost of a given level of AI performance has fallen ~47% per quarter since 2023, or ~13x per year. On one PhD-level science test, the same score went from ~30 cents a question to four-hundredths of a cent in under 18 months. Epoch’s comparison: a new car’s sticker price dropping from $50,000 to $69.

AI agents can handle longer jobs. METR, a research group that tests agents, tracks how long a task would take a human expert and whether an agent can finish it alone. Its January 2026 update puts the long-run doubling time at ~7 months, and ~3 months since 2024. The error bars are wide, but the direction isn’t: agents are moving from “draft this email” toward “work through this bid package.”

When building gets cheap, the bottleneck moves. Andrew Ng said on the No Priors podcast that work that once took six engineers three months, “my friends and I, we’ll just build on a weekend,” and “the bottleneck is deciding what do we actually want to build” (Business Insider). After workshops with ~100 CEOs and operators, Brian Solis wrote last month that the real bottleneck is imagination: most leaders still ask where to bolt AI on, not what they can now do that they couldn’t before.

Shop-floor version: for decades, office capacity was something you bought one salary at a time, like renting laser time from a service bureau. Each new estimator or PM was a fixed slot of capacity. AI agents and purpose-built software let you own that capacity instead, more like putting your own machine on the floor. Once you own it, adding the next job costs a lot less than adding the next person.

Can AI help you get into a new vertical?

Most fabricators have a market they’ve eyed for years: miscellaneous metals, data centre steel, energy and utility work, an adjacent product line. What usually kills it isn’t welding skill. It’s everything around the welding:

• learning a new spec world and its paperwork,
• quoting a new kind of package without wrecking your current backlog,
• building the submittals, compliance documents and tracking a new customer type expects.

That’s the kind of work AI now handles well. An agent can read a stack of new-vertical specs and flag what’s different from your usual jobs. A purpose-built quoting tool can encode the new rules so estimators aren’t starting from zero. Paperwork that used to need a dedicated hire becomes a workflow.

You still decide whether the vertical is worth it. AI just lowers the cost of a serious test from “hire two people and hope” to “try ten bids and see.”

Can you 2x growth without hiring a matching headcount?

Here’s the quiet cap in most mid-market shops: the PM team decides how many jobs you can take. Each project manager can carry only so many jobs before RFIs, change orders and delivery follow-up start slipping. Sales can find more work. The floor can often add a shift. Nobody finds the next good PM in a hurry.

That’s not a hunch. In 2022, a Canadian Manufacturers & Exporters survey of 563 manufacturers found 62% had lost or turned down contracts and faced production delays because they couldn’t find workers. In the US, Deloitte and The Manufacturing Institute estimate ~1.9 million manufacturing jobs could go unfilled by 2033. Hiring your way to 2x is a plan many shops can’t execute even with the money.

So leading manufacturers are aiming AI at growth. Manufacturers Alliance’s 2026 survey of 100+ mid- to large-cap manufacturers, The Great Acceleration, found that for many firms “the initial goal is simply growth with existing headcount,” and 48% count revenue growth when they measure AI’s return. One heavy industrial manufacturer told the researchers: “We don’t measure it because it doesn’t cost much, and I can’t hire fast enough to get the work done.”

This is the framing we use with clients: 2x the output, same team. Workflow automation and AI agents take on the coordination grind (chasing RFIs, updating schedules, drafting submittals) so each PM can carry more jobs well, at ~20% of the cost of a PM. For the mechanics, see what an AI agent actually is.

Can you launch new products or services you couldn’t staff before?

Think about the estimating backlog. Every shop has one. When estimators are buried in this week’s bids, the new product line never gets a price book, the new service never gets a quote template, and the idea quietly dies in a folder called “Someday.”

New-product launches rarely fail on the shop floor. They fail in the office, because nobody has time to work out the costing, quoting rules and follow-up. That’s the work AI and custom software now compress:

• Quoting a new product: a purpose-built tool that knows your materials, labour rates and margin rules can price a standard variant in minutes, and estimators review the edge cases. (More in our guide to using AI in a steel fabrication shop.)
• Offering a new service: inspection reports, as-built packages and maintenance programs all need paperwork. An agent drafts it, a person signs it.
• Selling to a new kind of customer: a portal for repeat buyers or a self-serve quote form used to be a big software project. With AI-assisted development, it’s a much smaller one.

Siemens’ Sabrina Joos summed up the mood in the Manufacturers Alliance report: AI “will expand what manufacturing organizations can do, not just automate what they already do today.”

What else is sitting on your “someday” list?

Every owner has one, even if it’s never been written down. Typical entries:

• “Bid the bigger jobs we always pass on because the package is 3,000 pages.”
• “Finally get one source of truth instead of three spreadsheets and a whiteboard.”
• “Replace the software we pay a fortune for and only use 20% of.”
• “Answer every quote request the same day.”
• “Get the owner out of the daily email firehose.”

Look at why each one stalled. It’s almost always one of three things: headcount (no one to do it), software cost (the tool was too expensive or didn’t fit), or coordination overhead (too many hand-offs). Those are exactly the three costs that are collapsing. Software is the clearest case: building around how your shop runs is now cheap enough that our purpose-built systems come in at ~10x lower annual cost than a big off-the-shelf fab suite (see pricing). Worried about where your drawings go? Our post on open vs closed AI models covers keeping data in-house.

What can’t AI do for your growth plan?

Time for the honest part.

• It won’t pick the strategy. AI can help you size a new vertical. It can’t tell you whether you want to be a data centre steel shop. Where you grow and which customers you turn away are still your calls.
• It won’t set your quality bar. A faster quote is worthless if it’s wrong. Someone with judgment still approves anything that touches price, schedule, safety or a stamp.
• Bad data and bad processes just fail faster. In the Manufacturers Alliance survey, 63% of manufacturers said their data needed clean-up before AI pilots could start. MSA Safety’s CIO put it well: “AI didn’t create these problems, it made them appear.”
• The bottleneck moves, it doesn’t vanish. As Ng wrote in April, when you speed up coding 10x or 100x, “everything else becomes slow in comparison.” Shops work the same way: free up the office and the next limit might be the paint line, a welding certification or cash flow on bigger jobs.
• AI doesn’t add machines. Workflow automation grows office capacity. If the floor is already maxed, growth also means equipment, people or partners.

None of that is a reason to wait. It’s a reason to aim AI at a specific growth goal instead of sprinkling it around.

How do you start? Fill in your five-question dream list

Block an hour this week. Pretend headcount, software cost and coordination aren’t limits. Write one honest answer to each:

1. New vertical: What market or customer type would we chase if quoting and paperwork for it were nearly free?
2. New product or service: What would we sell next if pricing it and documenting it took a day, not a month?
3. Capacity goal: How many more jobs a month would we take if each PM and estimator could carry 2x, and what does that do to revenue?
4. Process to free: Which single process eats the most skilled hours today and adds the least value? That’s your first automation candidate.
5. First small bet: What’s one experiment we could run in 30 days (ten bids in the new vertical, one new quote template, one automated follow-up loop) to prove it?

Then sort each answer: does it need strategy (you), software (build it) or workflow automation (an agent plus a person approving)? Most lists point back to the same one or two constraints.

Want help turning your dream list into a plan?

Embedding AI builds purpose-built software and AI workflow automation for mid-market fabricators and manufacturers, so the growth ideas you shelved for lack of people or budget get a real shot. Bring your dream list. Book a free 30-minute consultation and we’ll pick the one constraint holding you back, map how to remove it, and show you what ~2x with the same team could look like for your shop. Prefer email? Write to info@embeddingai.ca.

Sources

• Epoch AI, The plunging price of thought (September 22, 2026).
• METR, Time Horizon 1.1 (January 29, 2026).
• No Priors podcast, Ep. 128 with Andrew Ng (August 2025); Business Insider via Yahoo, Andrew Ng says the real bottleneck in AI startups isn’t coding (August 22, 2025).
• Andrew Ng, The Batch, AI-Native Software Development Needs Generalists (April 17, 2026).
• Brian Solis, What CEOs Say About AI in Private and What That Means To Leadership (September 9, 2026).
• Manufacturers Alliance, The Great Acceleration: Scaling AI from Tactical Pilots to Strategic Transformation (May 2026); press release (May 20, 2026).
• Canadian Manufacturers & Exporters, Canada leaving almost $13 billion on the table due to labour and skills shortages in manufacturing (October 25, 2022).
• Deloitte Insights, US manufacturing labor impact (citing Deloitte and The Manufacturing Institute).

Frequently asked questions

How can AI help my business grow, not just cut costs?

Point it at the work that caps how many jobs you can take: estimating, project coordination, submittals, follow-up and reporting. When that work stops needing a matching hire, the same team can quote more, run more jobs and try new markets. Growth comes from reinvesting the freed-up capacity, not from removing people.

Can a mid-market manufacturer really 2x output without hiring a matching headcount?

For the office side of the business, often yes. Workflow automation and AI agents can take on the repetitive coordination work that limits each estimator or PM. Shop-floor capacity is a separate question: AI does not add a press brake or a certified welder, so check that the floor can keep up too.

What is the best first step to scale manufacturing with AI?

Pick one constraint that is costing you real work today, such as an estimating backlog or a PM team at its limit. Map how that work flows, fix the obvious data mess, then automate one slice of it with a person approving the output. Measure jobs quoted or run per person before and after.

Do I need custom software to grow with AI?

Not always. Off-the-shelf tools cover common jobs. But if your growth plan depends on how your shop actually runs, such as a new product line, an unusual quoting process or a new vertical, purpose-built software has become far cheaper to build than it was even two years ago.

What can't AI do for a growth plan?

It cannot choose your market, set your quality bar or decide which customers to say no to. It also speeds up whatever process you hand it, so bad data or a broken process just fails faster. Owners still own the strategy; AI removes the excuse that there were never enough people to try it.