David Shijoon Bae and Michael Lin started building on April 18, 2026. A month later, the two Harvard dropouts had landed $500,000 from Y Combinator. In April they launched FlowManual, an AI platform designed to automate the kind of back-office work that construction contractors would rather not think about—bidding, purchasing, contract review, invoice reconciliation.
It's unglamorous stuff. Yet for a general contractor juggling dozens of subcontractor bids, each laden with exclusions and exceptions, the hours pile up fast. FlowManual emerged from Y Combinator's S26 batch with a simple premise: most of this comparing, flagging, and reconciling can be done by software that actually reads the documents.
The pitch is straightforward, if ambitious for a two-person team. Construction contractors spend entire afternoons comparing bids to contracts, hunting for scope changes buried six pages deep in a PDF, then reconciling invoices against purchase orders to catch the line items that somehow weren't in the original buyout. FlowManual ingests those documents and surfaces the differences that matter—flagged on the exact clause, down to the section number.
Clause by clause, line by line
FlowManual describes itself as "the back office for construction contractors, from the bid to the buyout." The platform pulls in bids, contracts, vendor quotes, and invoices, then applies AI to highlight scope changes, price shifts, exclusions, and deadline adjustments. Each discrepancy comes with a direct reference to the source: §7.4(b), for instance, or a specific invoice line.
The intelligence layer divides neatly into two phases. Pre-award, the system runs a side-by-side comparison of bid versus contract, ranking differences by severity and potential dollar impact. It flags when scope has crept in, when exclusions have been quietly dropped, when insurance terms or schedules have shifted. Post-award, it reconciles invoices against the locked buyout figures, catching price drift and items that were never in the purchase order to begin with.
A central "workbench" pulls together AI-assisted takeoffs—users can accept or reject the pins the system drops on plan sheets—along with revision overlays, first-pass estimates, an RFI log, submittal register, and change order tracking. The company's website shows UI snippets of invoice-versus-buyout comparisons: an RTU price up $3,200, clause references, the sort of granular detail that might save a project manager an hour of cross-checking spreadsheets.
FlowManual also maintains a historical purchase order price book with a 90-day outlook, computing what it calls "lock-or-wait" flags to help contractors time their material buys. Whether contractors will trust those signals enough to act on them remains to be seen.
Two founders, tight budgets, broad ambitions
Bae studied mechanical engineering at Harvard before dropping out; Lin previously interned as a software engineer at Dijie and co-authored a research paper submitted to an IEEE conference while at UC Santa Cruz. The team remains just the two of them, according to Y Combinator's directory and LinkedIn, which lists FlowManual in the 2–10 employee range—startup shorthand for "very small."
The company took Y Combinator's standard deal: $125,000 for 7% equity on a post-money SAFE, plus $375,000 on an uncapped most-favored-nation SAFE. Bae confirmed the $500,000 total in a LinkedIn post not long after the raise closed.
On the technical side, FlowManual uses Anthropic's models for text analysis with zero retention and no training on customer data unless a user explicitly opts in. Microsoft Azure handles hosting and encrypted file storage; Azure Document Intelligence parses PDFs but scrubs the analysis data within 24 hours. The security page lays out encryption in transit (TLS 1.2+) and at rest (AES-256-GCM), item-level key derivation with HKDF-SHA-256, and HMAC blind indexes for searchable fields. Passwords are hashed with scrypt; two-step email verification is optional; brute-force lockouts are standard. SOC 2 Type II certification is in progress, and an independent penetration test is on the calendar.
The company stores no ad-tracking cookies and relies on PostHog for privacy-preserving product analytics—metadata only, no filenames or vendor names logged. For a seed-stage startup handling sensitive contract data, that's table stakes, not table scraps.
A category getting crowded
Construction software is fragmenting, fast. Procore announced expanded AI agents for submittals, RFIs, and contract reviews earlier this year, embedding its Datagrid platform across core workflows. Trimble agreed to acquire Document Crunch in early spring, signaling that contract intelligence is no longer a nice-to-have—the deal, expected to close later this year, will fold Document Crunch's risk platform into Trimble Construction One.
Other players occupy adjacent slices of the workflow. Togal.AI focuses on AI-assisted takeoff, claiming accuracy rates nearing 98%. PlanHub and BuildingConnected (owned by Autodesk) run bid-management networks connecting generals, subs, and suppliers. STACK positions itself as an "AI-accelerated takeoff and estimating" platform. Fresco, another recent Y Combinator graduate, is building an AI copilot for estimators with a focus on Division 8 work—doors, windows, glass.
FlowManual's pitch is that it bridges preconstruction and post-award procurement on a single workbench. Where Document Crunch emphasizes contract risk intelligence and Procore embeds agents across many workflows, FlowManual explicitly reads both the bid and the contract—then follows through to invoice reconciliation. The company's terms of service describe features including automated document annotation, contract-to-bid comparison, scope gap detection, vendor price tracking, and submittal review, with an explicit disclaimer that AI output can make mistakes and requires independent verification. Sensible, if not exactly confidence-inspiring.
Free to start, custom to scale
FlowManual offers a free plan with all features, daily usage limits, unlimited projects and uploads, side-by-side comparison, estimates, chat, and email support. The Pro plan is sales-led with custom pricing, unlimited usage, and priority support. No public seat-based pricing is posted—common for enterprise software, less so for a product that also offers a free tier. Both plans run on FlowManual's hosted cloud; enterprise deployment options are mentioned but not detailed.
The company's website went live with product screenshots and dual calls to action: "Start free" and "Book a 20-minute demo." That suggests the platform is in open beta or general availability, though it's not entirely clear. LinkedIn company posts from recent weeks include short demo videos and positioning language emphasizing that the tool works "alongside your team (not replace)"—a reassurance that's become standard in AI product marketing.
No customer logos or case studies appear on the site or LinkedIn as of now. VibeCrowd, a launch aggregator, added a FlowManual entry in early July, noting the company's Y Combinator participation. No mainstream press releases or independent product reviews have surfaced yet, which isn't surprising given the company's timeline.
The bet
Industry signals suggest AI adoption in construction is accelerating, though perhaps unevenly. The Associated General Contractors' outlook earlier this year highlighted rising AI investment, with preconstruction and design among the use cases. An ASCE Journal systematic review published in June points to a concentration of AI applications in contract document analysis and risk management. Buildr's recent roundup emphasizes AI for document review, bid leveling, and staffing recommendations in the CRM preconstruction stages.
Trimble's acquisition of Document Crunch drew a telling observation from Engineering News-Record: contract intelligence is becoming table stakes. FlowManual is betting that the category extends beyond risk flagging to the operational grunt work—matching bids to contracts, leveling subs after normalizing exclusions, catching invoice line items that weren't in the buyout.
For a two-person team a few months into building, that's a lot of ground to cover. The product surface is broad: takeoffs, bid leveling, purchasing intelligence, submittals, RFIs, change orders, invoice reconciliation. The free plan lowers the barrier to trial, but scaling will require either rapid headcount growth or ruthless focus on the workflows that contractors will actually pay to automate. Or both.
The next inflection point is Demo Day in September. Until then, FlowManual remains a pitch, a live product with no public customers, and a bet that construction contractors are ready to hand AI the keys to the back office—provided it shows its work, clause by clause, line by line. Whether that's enough to stand out in a crowded field is the question Bae and Lin will spend the next several months answering.
