By Sagar Shankaran, Founder of CallSphere
One estimator, 38 RFQ packets a week, half of them never quoted. What Claude Cowork and ChatGPT Work change about aerospace build-to-print quoting in 2026.
Key takeaways
The estimator gets in before anyone else because that is the only quiet hour in the building. He opens Exostar for the Boeing-side packages, then the prime's own supplier portal for the Northrop work, then the sales inbox where second- and third-tier requests arrive as plain email with a zip file attached. Nine new request-for-quote packets since Friday. Each is a drawing in PDF, a bill of material, two pages of quality clauses, a note that the order will carry a DO-A7 rating under the Defense Priorities and Allocations System, and thirty-odd pages of terms nobody has read end to end since 2019.
By noon he will have priced two. Two more go out Thursday. The rest get no-quoted, and the no-quote is not a decision anybody made — the calendar made it. Every build-to-print owner knows that number and hates it: the bids you never sent.
Walk the job honestly and you see where the hours go. Read the notes block, find every specification callout: AMS 2759 for heat treat, AMS 2700 for passivation, MIL-DTL-5541 Type II Class 3 for the chem film. Check whether your outside processor's NADCAP accreditation actually covers that process at that class — not the one next to it on the scope sheet. Build a routing in Epicor Kinetic or JobBOSS² or E2 with setup hours, cycle times, in-process inspection and the AS9102 first article nobody ever costs properly. Email the metal house for a price on 15-5PH bar and wait a day. Email the anodizer. Then apply your burden rate. Two and a half hours on a simple machined detail; a day and a half on a small assembly with six purchased components and a torque-and-lockwire operation.
The distortion this creates is worse than the hours. You quote what is easy to quote: repeat part numbers get priced Tuesday, and the unfamiliar package with two special processes sits until it expires. Your mix drifts toward whatever your estimator can price fastest, which is not the same thing as whatever your shop makes money on.
The 2026 change worth an owner's attention is not that a chat window can read a drawing note. It is that you can now hand over the whole errand. An agent that hands back finished work is software you give a goal to — "price this package by Thursday" — which then opens your files and systems, does the steps in order, and puts a finished quote packet on your desk, instead of answering one question at a time and waiting for you to think of the next one.
Claude Cowork landed on 12 January 2026 and reached web and mobile in July. ChatGPT Work followed on 9 July 2026, running on GPT-5.6. Same idea from two directions: you state a goal, connect it to your files and applications, it works on its own for a stretch — sometimes hours — and hands back a finished artifact. Both were aimed squarely at people who do not write software, which in a machine shop means the estimator, the buyer, the quality manager and the contracts administrator.
The 2024 version could summarize a drawing note if you pasted it in. The 2026 version opens the drawing PDF, the spec list, your process capability sheet and your last four quotes for that customer, and produces the routing and the cost sheet in the format your ERP wants, plus a page of exceptions it wants you to look at. The practical proof is that the output is a document, not a paragraph.
Hear it before you finish reading
Talk to a live CallSphere AI voice agent in your browser — 60 seconds, no signup.
flowchart TD
A["RFQ packet lands in the supplier portal"] --> B["Agent reads drawing notes and spec callouts"]
B --> C["Agent drafts routing in the ERP format"]
C --> D{"Every process covered by a current NADCAP source?"}
D -->|No| E["Agent flags the gap, drafts the outside-processor RFQ"]
D -->|Yes| F["Agent pulls material price and builds the cost sheet"]
E --> F
F --> G["Estimator reviews, sets margin, releases the quote"]
The estimator does not open nine packets. He selects all nine and writes one instruction: price these against our standard burden, use current metal quotes if under fourteen days old, flag anything needing a process we are not approved for, and list every flowdown clause that differs from what we accepted last time from this customer. Then he goes to the production meeting.
By late morning there are nine draft packets. Each carries a routing he can read, a material line with the distributor and the date of the price, a first-article line with hours in it, and — the part that matters most — a short exceptions page. Package four: the note block calls for shot peen to AMS 2430 and your approved peening source lapsed in April. Package seven: the terms require source inspection at final where the last three orders from that customer did not, which changes inspection hours. Package nine: the drawing is one revision behind the revision named in the purchase order text.
That last one is the whole argument. A junior estimator misses it, you cut chips to revision H, and you find out at first article that the customer bought revision J — a corrective action and a no-charge rework on a job you underbid.
What the owner has to change is the harder half. You stop assigning tasks and start writing definitions of done. "Quote this" is not a goal; it is a wish. Write it down once: every spec callout with its approved source and accreditation date, material priced within fourteen days, first article hours included, exceptions listed separately, nothing sent to a customer without a human signature.
Illustrative assumptions, not a case study. Substitute your own quote log from last quarter.
| Assumption | Today | With the agent drafting |
| RFQ packets received per week | 38 | 38 |
| Packets actually quoted | 22 | 32 |
| Estimator hours per quote | 2.6 | 0.8 (review and price only) |
| Estimator hours per week | 57.2 | 25.6 |
| Loaded estimator cost at $52/hour | $2,974 | $1,331 |
The labor line saves about $1,643 a week, which is real and is not the point. The point is the ten extra packets. Assume a conservative 12% hit rate on those marginal bids — lower than your house average, because these are the unfamiliar ones — an $8,400 average job and a 26% gross margin. Ten extra quotes at 12% is 1.2 jobs, roughly $10,080 in revenue and about $2,621 in gross margin per week, against a seat cost in the low hundreds per month.
Prove it the way you prove a process change: keep an eight-week quote log with three columns — received, quoted, won — and compare it to the same eight weeks last year. The number you are testing is whether more bids left the building.
This is where an aerospace supplier is different from a sign shop, and it is not optional. If the drawing is technical data controlled under the International Traffic in Arms Regulations, putting it into a general-purpose service is an export the moment a foreign person can see it. If the contract carries DFARS 252.204-7012, the drawing is controlled unclassified information and has to sit somewhere that meets NIST SP 800-171 — and your CMMC Level 2 assessment will ask exactly where it sat.
Still reading? Stop comparing — try CallSphere live.
CallSphere ships complete AI voice agents per industry — 14 tools for healthcare, 10 agents for real estate, 4 specialists for salons. See how it actually handles a call before you book a demo.
Practical answer: start on commercial work. Quote the non-controlled packages first, get your definition of done stable, and keep controlled packages on the manual route until the vendor gives you a written answer about where the data is processed and stored, in a form your export compliance officer and your assessor accept. Anthropic's 2 July 2026 governance update — spend limits, alerts, per-user entitlements — matters here for the reason people miss: it records who was allowed to use what, which is what an assessor asks to see.
The margin call does not transfer. Whether you bid this customer at 24% or 31% is a relationship judgment: how they pay, whether they short-run you, whether this part opens a family of part numbers, whether the government fiscal year closing on 30 September means they need you badly in eight weeks.
Neither does capacity truth. The agent will happily produce a routing needing your Swiss lathe for eleven days in a window already committed. Your shop foreman knows what is really on the floor; the finite schedule in the ERP usually does not.
And neither do the ugly parts — thin-wall titanium with true position on a bore you can reach one way, or a weldment that warps at heat treat. That memory lives in your programmer and your material review board history. Feed it in as your own notes and the drafts improve; do not expect it off the drawing.
Assume you export and import for now. The realistic 2026 setup is a routing and cost sheet laid out exactly like your ERP import, which one person checks and loads. Ask your ERP reseller what their agent connection looks like before you buy on the strength of a demo.
It reads it better than it used to and still needs a human on the revision letter. Treat any hand-marked or scanned drawing as an automatic exception the estimator opens personally. That rule costs a few minutes and saves the rework job.
No. Keep a human signature on anything that leaves the building with a price on it. A quote is an offer, and an offer with a wrong burden rate on a five-year agreement is an expensive typing error.
Take the twelve packages you no-quoted last month — already dead, so they cost you nothing — and give the whole batch to the agent as one goal. Compare the drafts to what your estimator would have done. You will learn more in an afternoon than in three vendor demonstrations, without risking a live bid.
One more thing about quoting season: every extra bid generates phone calls. Buyers confirming you got the package, expediters chasing status the week before the government fiscal year closes, an AOG call at 6:40 p.m. that goes to voicemail. CallSphere builds AI voice and chat agents that answer the phone line and web chat around the clock, take the part number and need-by date, and put the message in front of the right person. It does not price your work — your estimator does that — but it makes sure the request reaches him.

Written by
Sagar Shankaran· Founder, CallSphere
LinkedInSagar Shankaran is the founder of CallSphere, where he builds production AI voice and chat agents deployed across healthcare, hospitality, real estate, and home services. He writes about agentic AI, LLM engineering, and shipping voice agents that handle real calls in production.
See how AI voice agents work for your industry. Live demo available -- no signup required.
The monthly IEEE 1366 reliability close takes 64 hours across three people. What goal-driven agents change, the arithmetic, and what stays with the engineer.
Seasonal spend ceilings, per-person limits, 75% alerts and cost per quote packet: budgeting AI at a contract assembly shop without stalling the quoting desk.
The irreversible actions in a precision machining shop that must keep a human in the loop, and how to scope everything else an AI assistant touches in 2026.
How pest control service managers hand the monthly food-account trend packet to a 2026 work agent as a goal - and what has to change about assigning work.
The phased plan, insurance estimate, predetermination narrative and financing page, finished before the patient leaves. What the owner has to change to get it.
Why co-pack quotes take six days, and how 2026 agents that return finished work rebuild the packet — costed formula, freight, spec sheet — in two hours.
© 2026 CallSphere Inc. All rights reserved.
Made within San Francisco
Watch how CallSphere handles real customer calls, schedules appointments, and processes payments — live.
Try Live DemoBook a DemoCalculate Your ROI