
In Could, two researchers on the College of Southern California printed a paper describing a tool I have wanted for fifteen years: a conversational AI agent that pulls 1000’s of fragmented federal and basis grant portals into one place, reads your draft proposal, and surfaces the alternatives that match.
The authors report it reducing grant discovery from the thirty minutes of handbook portal clicking down to 10. Prior to now, I’d have learn it, admired it, and filed it underneath “issues somebody with a funds ought to construct sometime.” As an alternative I described what I needed to Claude Cowork and by the tip of the afternoon I had a working model.
Funder Finder: takes your focus areas or generic proposal PDF and it queries a unified index of roughly 12,000 federal and basis alternatives, falling again to reside internet search just for the previous couple of days, with each consequence linked to the true company or funder web page.
A New Approach to Create Software program
A non-engineer turning a peer-reviewed techniques paper right into a working internet app between lunch and dinner would have been a fantasy two years in the past. It’s bizarre now, and the event sector wants to just accept this new actuality.
The reflex in our sector continues to be that actual software program requires grant funding and a procurement cycle. You persuade management, write a proposal, anticipate a win, then rent a guide, write an RFP, consider distributors, and a 12 months and a quarter-million {dollars} later you will have a platform.
That reflex constructed a complete class of GenAI grant-writing instruments to search out alternatives and draft compliant proposals. The reflex made sense when constructing software program meant using dozens of software program engineers. It now not describes how a rising share of helpful software program will get made.
Sign Up Now for extra software program engineering for good insights
Constructed by Person
My pal Alexis Bonnell calls this strategy: Constructed by Person. A site skilled who is aware of the ache, is aware of what success seems to be like, and is accountable for the consequence builds the software program themselves, in days, for the worth of an LLM subscription.
This isn’t low-code. The earlier technology of “democratize software program” instruments, Bubble, Airtable, Zapier, gave you pre-built blocks to snap collectively, and the consequence was at all times constrained by what the seller imagined. They continue to be highly effective for normal workflows, which is why Gartner tasks the mixed low-code and no-code market will reach $44.5 billion in 2026. However when your downside is particular sufficient that no vendor has imagined it, low-code hits a wall.
AI coding brokers work in another way.
Claude Code and OpenAI Codex learn a codebase, write and edit recordsdata, run assessments, and produce commonplace code you personal and may deploy anyplace. The output is bounded by your understanding of the issue, which for a site skilled is commonly deeper than any vendor’s.
You additionally inherit the upkeep, the debugging, and the safety threat that comes with actual code, some extent I’ll return to as a result of it’s the place many of the sector’s optimism is misplaced.
Constructed by Person: Me

The proof for Constructed by Person is embarrassingly literal for me. I had by no means opened GitHub earlier than Could 2026. Matt Berg and Richard Stanley can affirm my lack of software program engineering expertise.
A month in, I now have 88 contributions across ten public repositories, most of them instruments for NGO fundraising, grant compliance, and digital well being constructed with Cowork and Codex. For instance:
Commit rely is a crude measure and most of these commits are small. The form is the purpose: from zero to a working portfolio in a month, by somebody who couldn’t have coded his means out of a paper bag in 2025.
The Instrument I Care About Most
In 2023, researchers at USC’s Mind Well being Observatory analyzed Medicare knowledge covering 40 million Americans over 65 and located that solely about 8% of anticipated delicate cognitive impairment instances had been recognized, leaving an estimated 7.4 million cases undetected. A companion examine discovered that main care clinicians underdiagnose MCI in roughly 99% of cases.
The explanations are structural: fifteen-minute visits, weak reimbursement, no commonplace protocol for what occurs after a constructive display screen. In the meantime the 2 FDA-approved Alzheimer’s medication, lecanemab and donanemab, solely assist sufferers nonetheless within the early window. Miss the window and the affected person is now not eligible.
This issues to me as a result of my mom has dementia. She was recognized too late for both drug.
No product available on the market lets a household caregiver run a validated screening protocol at dwelling and stroll into a physician’s workplace with a clinician-ready packet particular sufficient to set off a referral.
So I constructed one.

First Signs Health combines 4 validated cognitive devices in a three-gate browser protocol. It doesn’t diagnose. It produces a handoff doc a doctor can act on. I constructed it in days, for about $100 a month in instruments, as a non-engineer who turned a useful skilled in dementia as a result of my household pressured me to.
Constructed by Person in Software program Improvement
The non-public sector market has already reacted to Constructed by Person. In late January 2026, Anthropic launched industry-specific plugins for its Claude Cowork agent. By February:
Attributing inventory strikes to single product launches is unreliable, and these figures span promoting, consulting, and monetary knowledge nicely past SaaS. The path, although, is just not doubtful.
SaaS is just not dying. It’s splitting.
Single-purpose instruments whose worth is organizing, looking, or synthesizing info are probably the most uncovered. Instruments whose worth is maintained knowledge infrastructure, workflow orchestration throughout complicated techniques, or regulatory compliance have way more runway.
No vendor was ever going to construct First Indicators Well being – there is no such thing as a worthwhile marketplace for it right now – and a software that handles HIPAA-compliant storage, integrates with an digital medical report, and maintains audit trails nonetheless requires enterprise engineering.
Constructed by Person doesn’t threaten complicated SaaS. It might make it extra needed, as a result of it places extra half-built software program into the world that finally wants to hook up with actual infrastructure, and actual infrastructure continues to be any person’s full-time job.
We Want Extra Software program Engineers
The optimism I most mistrust is the concept Cowork and Codex displaces engineers. It won’t. It’ll change what they do.
The Bureau of Labor Statistics projects 15% employment growth for software program builders, QA analysts, and testers from 2024 to 2034, a lot quicker than common, and CNN reported in April 2026 that developer openings are rising exactly as a result of non-engineers can now prototype, growing demand for individuals who can end the job safely.
The reason being failure modes.
Veracode’s 2025 GenAI Code Security Report examined greater than 100 massive language fashions throughout 80 duties and located that AI-generated code launched safety vulnerabilities in 45% of instances, with no enchancment from newer or bigger fashions. A printed screening guidelines fails visibly. Software program fails invisibly.
My dementia software makes use of printed scoring protocols, however a poorly-built screener may miscalculate a rating and reassure somebody who needs to be seeing a neurologist. The engineering judgment that catches that’s changing into extra beneficial, not much less.
Constructed by Person Future in ICT4D
1. Please cease treating a $250,000 customized platform because the default.
A basis about to fund one ought to first ask whether or not a program officer, working with a coding agent for per week, may prototype the core performance. That experiment will reveal which 20% of the construct truly requires skilled engineering, and it’ll expose how a lot the price construction of social-sector know-how has already modified.
2. Create a ruled Constructed by Person course of.
Let area consultants construct and deploy the straightforward issues, no delicate knowledge, no authentication, no manufacturing integration, with specific guardrails, and pay engineers to deal with all the pieces that touches actual knowledge or actual choices.
Funders that finance digital instruments ought to ask grantees a blunt query they can’t presently reply: who maintains this, and who reviewed the code. I’ve argued earlier than that extra proposals written with AI is just not the identical as extra good proposals, and the identical warning applies right here.
3. Extra software program is just not the identical as extra working software program.
The dementia screening hole won’t be closed by a fifteen-minute main care go to, and it’ll not be closed by a SaaS product constructed for the most important addressable market. Will probably be closed by folks like me who see the issue up shut and now have the instruments to behave.
Whether or not these instruments are protected sufficient to belief will depend on whether or not we pair area experience with engineering judgment, as an alternative of creating one wait in line behind the opposite. The software program is now a budget half. The judgment is just not.









