# SI (Super Intelligence) Development Companies: How to Choose One

URL: https://vrisic.com/blog/super-intelligence-si-development-companies-how-to-choose/
Last updated: 2026-10-18

**Short answer:** “SI” is the federal naming for the technology previously called AI. Executive Order 14434 (29 September 2026) defines Super Intelligence as the technologies covered by the existing statutory definition of artificial intelligence, so an SI development company builds the same software as an AI development company. The order binds federal agencies, not private companies. Choose on what the software does on your data, how it is tested, who owns it and what it costs, and treat any vendor selling a system beyond human ability as a red flag.

In this guide

1. What “SI” means in the order
2. The two meanings vendors mix up
3. What a vendor can honestly sell
4. Who needs SI wording
5. What federal buyers actually ask vendors for
6. Ten questions to ask
7. A weighted scoring sheet, with an example
8. Red flags in SI marketing
9. Contract terms to require
10. What SI software costs
11. Three buyer scenarios
12. When not to hire a development company
13. A low-risk selection process
14. Where to start your shortlist
15. How Vrisic can help
16. Sources

Since the order, buyers have started searching for SI development companies, SI software and Super Intelligence consultants. Many results are ordinary AI vendors with a new label, and a few are making claims no one can back up. This guide helps a US buyer, or a company selling to US buyers, separate real capability from relabelling. For the order itself, read our explainer on [AI vs SI and what the US order means](https://vrisic.com/blog/ai-vs-si-super-intelligence-us-businesses/) .

**Disclosure:** Vrisic wrote this page and sells AI software, so we have a commercial interest. We have no product beyond human ability and we say so below. Verify the order text yourself on whitehouse.gov; links are in the sources. This is not legal advice.

## What does “SI” mean in the executive order?

In the order, “Super Intelligence” and “SI” are a naming convention for the same technology the law already calls artificial intelligence. Section 3(a) of Executive Order 14434, “Inaugurating the Era of Super Intelligence”, signed on 29 September 2026, says the terms mean the technologies and systems covered by “artificial intelligence” as defined in 15 U.S.C. § 9401(3). That statutory definition comes from the National Artificial Intelligence Initiative Act of 2020 and describes a machine-based system that, for human-defined objectives, makes predictions, recommendations or decisions ([White House, order text](https://www.whitehouse.gov/presidential-actions/2026/09/inaugurating-the-era-of-super-intelligence/) ; [15 U.S.C. chapter 119](https://www.govinfo.gov/content/pkg/USCODE-2020-title15/html/USCODE-2020-title15-chap119.htm) ).

The order is also short on obligations for anyone outside government. Here is what its text provides, based on the whitehouse.gov version and the Federal Register public inspection copy ([Federal Register, document 2026-20321](https://public-inspection.federalregister.gov/2026-20321.pdf) ):

| Section | What it says |
| --- | --- |
| 2(a) | Executive departments and agencies must, to the maximum extent permitted by law, use “Super Intelligence” and “SI” in place of “Artificial Intelligence” and “AI” in official correspondence, public communications, websites, reports, policy documents and other non-statutory documents |
| 2(b) | Previously issued regulations, presidential actions, contracts, grants and other historical documents do not need to be changed |
| 3(a) | SI means the technologies covered by the statutory definition of AI, unless and until superseded by later presidential action or an Act of Congress |
| 3(b) | Within 60 days the Assistant to the President for Science and Technology must submit proposed legislative language for a federal definition of SI, with an assessment of whether it should modify or supersede the AI definition and proposed conforming amendments |
| 4 | The order is subject to applicable law and appropriations and creates no enforceable rights or benefits for any party |

Sixty days from 29 September lands on 28 November 2026 by our calculation, so the definition proposal may be public soon after you read this. The order text does not mention private companies, contractors or AI developers. The accompanying fact sheet describes federal initiatives, including a Genesis Mission with more than $5 billion announced for science and a July 2025 action plan with more than 90 federal actions, and it does not name any company or voluntary industry commitment ([White House fact sheet](https://www.whitehouse.gov/fact-sheets/2026/09/fact-sheet-president-donald-j-trump-inaugurates-the-era-of-super-intelligence/) ).

Check the order page for later updates. If Congress or a later presidential action changes the definition, anything in this guide that depends on Section 3 may be out of date.

SI in the order is a naming rule, not new technology. Source: Executive Order 14434, whitehouse.gov and Federal Register, September to October 2026

## What are the two meanings of SI that vendors mix up?

There is the federal meaning above, and the older technical meaning of superintelligence. They are not the same, and some vendors blur them on purpose.

| Term | Meaning | Exists today? |
| --- | --- | --- |
| SI in EO 14434 | A federal name for the technologies covered by the statutory AI definition | Yes: it is the same software and models sold today |
| AGI | A system with roughly human-level ability across most intellectual tasks | No agreed example |
| Superintelligence (technical) | In Bostrom’s widely cited wording, any intellect that greatly exceeds human cognitive performance in virtually all domains of interest | No. A hypothetical; even firms pursuing it do not sell it ([Wikipedia](https://en.wikipedia.org/wiki/Superintelligence) ) |

The practical test: if a vendor’s SI claim would still be true under the federal definition, it is just AI software. If it would only be true under the technical definition, it is not for sale.

## What can an SI development company honestly sell?

Anything that is AI software today, which covers a lot. Typical deliverables:

- [AI agents](https://vrisic.com/services/ai-agent-development/) that read a request, look up data, take an action in your systems and escalate when unsure.
- [Voice assistants](https://vrisic.com/services/voice-ai-development/) that answer and route calls.
- Document and invoice processing, including extraction, matching and exception handling.
- [Search over your own files](https://vrisic.com/services/enterprise-rag-ai-search/) with cited answers.
- [Integrations](https://vrisic.com/services/ai-integration/) between models and your CRM, ERP or ticketing system.
- Evaluation, monitoring and governance around all of the above.

If a vendor sells these under the SI label, the price should not rise because of the label. If it describes something that sounds beyond all of this, ask for a demo, a test result and a written definition.

## Who actually needs SI wording?

| Buyer or seller | Need SI wording? | Reason |
| --- | --- | --- |
| Federal agency | Yes, in its own documents, to the extent the order says | Section 2 applies to executive departments and agencies |
| Contractor or grant applicant to a federal agency | Probably, in responses; follow each solicitation | The order does not name contractors, but your buyer’s documents may use the term |
| State and local government | Check its own style guide | The order is directed at the federal executive branch |
| Private company selling to private buyers | No requirement | The order creates no duties for private parties; customers still search for AI |
| Anyone citing a law or standard by name | No: use its exact title | The order does not rewrite statutes or existing documents |

A sensible default for a vendor website is to cover both terms side by side, lead with outcomes, and follow the buyer’s wording once you know it.

## What do federal buyers actually ask vendors for?

Federal AI buying is driven by procurement rules, not by the label. The Office of Management and Budget’s acquisition memo M-25-22 is a good guide to what contract terms look like. As summarised by a government contracts firm, it expects agencies to address government data use and IP rights, bars vendors from using non-public agency data to train commercial AI without consent, calls for terms on knowledge transfer, data and model portability and pricing transparency to avoid lock-in, allows solicitations to require disclosure of AI use during performance, directs agencies to maximise use of AI developed and produced in the United States, and requires ongoing testing and monitoring of performance and risk ([PilieroMazza summary of OMB M-25-22](https://www.pilieromazza.com/omb-issues-memoranda-on-use-and-acquisition-of-ai-by-federal-agencies-part-2-what-it-means-for-government-contractors/) ). The memo was written in the AI vocabulary; check whether current solicitations restate it as SI.

Another common reference is the NIST AI Risk Management Framework, released on 26 January 2023 as a voluntary resource for organisations that design, develop, deploy or use AI systems ([NIST AI 100-1](https://www.nist.gov/publications/artificial-intelligence-risk-management-framework-ai-rmf-10) ). Even private buyers use it as a shared vocabulary for risk questions.

The lesson for any buyer, federal or not, is that the useful vendor questions are about data, portability, testing and monitoring. They are the same under either label, and a vendor who answers them well is better evidence than one who uses the newer term.

Clear terms can outscore a slick demo. Vrisic illustrative example; weights: evidence 25, scope 20, ownership 20, price 15, security 10, support 10

## Which ten questions should you ask any SI or AI development company?

1. **What exactly will the software do on day one?** Ask for a one-page description of inputs, outputs and what a person still does.
2. **Which models and which cloud will it use?** And can you swap the model later without a rewrite?
3. **How do you test accuracy, and on whose data?** A good answer names a test set built from your real cases, the metrics and the target.
4. **What is the fixed price, or the cap, for the first release?** And what is excluded?
5. **Who owns the code, prompts, test data and model keys?** You should, in writing.
6. **What happens to my data and who can see it?** Where is it stored, is it used for training, how is it deleted?
7. **How are mistakes and human review handled?** Which actions need approval, and what is logged?
8. **What will it cost to run each month?** Hosting, model usage, monitoring, support.
9. **How do you define SI, and will you accept my definition?** If a solicitation defines it, the vendor should use that wording.
10. **Can I see a working demo on data like mine, and speak to someone who has used your work?** Newer firms may have few references; then ask for a small paid pilot.

## How do you compare vendors with a weighted scoring sheet?

Score each vendor from 0 to 3 on six criteria, multiply by a weight that reflects your priorities, and compare totals. The SI label earns no points by itself. These weights suit a typical business build; change them if, for example, security matters more.

| Criterion | Weight | 3 points means | 0 points means |
| --- | --- | --- | --- |
| Scope clarity | 20 | Written scope with exclusions and acceptance criteria | “We will figure it out as we go” |
| Evidence on your data | 25 | Demo or pilot results on your real examples | Slides only |
| Ownership terms | 20 | You own code, prompts, test data; runs in your accounts | Vendor-hosted, licensed access only |
| Price clarity | 15 | Fixed price or hard cap plus a stated running cost | Open-ended hourly with no cap |
| Security and data handling | 10 | Written controls, data stays in your cloud, no training on your data | No answer |
| Support and handover | 10 | Warranty period, documentation, handover session | None mentioned |

Formula: score = sum of (weight x points / 3), out of 100. Here is an illustrative example with three made-up vendors, not real companies.

| Criterion (weight) | Vendor A | Vendor B | Vendor C |
| --- | --- | --- | --- |
| Scope clarity (20) | 3 = 20.0 | 1 = 6.7 | 2 = 13.3 |
| Evidence on your data (25) | 1 = 8.3 | 3 = 25.0 | 2 = 16.7 |
| Ownership terms (20) | 3 = 20.0 | 1 = 6.7 | 2 = 13.3 |
| Price clarity (15) | 3 = 15.0 | 1 = 5.0 | 2 = 10.0 |
| Security (10) | 2 = 6.7 | 3 = 10.0 | 1 = 3.3 |
| Support (10) | 2 = 6.7 | 3 = 10.0 | 1 = 3.3 |
| **Total** | **76.7** | **63.3** | **60.0** |

Vendor B had the most polished demo and the strongest security story but would not accept ownership or price terms, so it ranks below Vendor A, which had clear terms but had not yet shown results on your data. The sensible next step is a short paid pilot with Vendor A to close the evidence gap. Scores are a conversation tool, not a verdict: if two vendors are close, let the pilot decide.

## What are the red flags in SI marketing?

- **Claims of human-beating intelligence.** Anything implying the product outperforms people across the board is unverifiable and, under the federal definition, not what SI means.
- **The label as the differentiator.** “Our SI platform” with no description of what changed in the software since last year.
- **Implied government endorsement.** The order does not certify or approve any vendor or product. Treat statements like “SI compliant” or “aligned with the White House order” as meaningless unless the vendor can point to a specific requirement.
- **Price premium for SI.** The same scope should cost the same under either name.
- **No definition.** A vendor that cannot say in one sentence what it means by SI.
- **Black box ownership.** Proprietary models you can never move away from, with no export of prompts, data or configurations.
- **Guaranteed accuracy.** Ask for measured accuracy on your data with confidence ranges, and what happens when the system is wrong.
- **A shiny rebrand and nothing else.** Old case studies with “AI” find-and-replaced to “SI”.

## Which contract terms should you require?

| Topic | What to write down |
| --- | --- |
| Definition | “AI or SI means the technologies covered by 15 U.S.C. § 9401(3), unless the buyer’s solicitation or a later law defines it otherwise.” This ties the term to something real and survives any renaming |
| Scope and acceptance | Written deliverables and measurable targets on your test set |
| Ownership | Assignment of code, prompts, evaluation sets and configuration to you |
| Hosting and keys | Runs in your cloud account; model API keys in your name |
| Data use | No use of your non-public data to train models; stated retention and deletion |
| Portability | Export of data, prompts and configuration; documentation and handover session |
| Monitoring | Agreed metrics, review cadence and incident process |
| Disclosure | Notice of material changes, such as swapping the underlying model |
| Payment | Milestones tied to working software |

Several of these mirror the themes in federal acquisition guidance above (data rights, portability, transparency, monitoring), which is deliberate: they protect private buyers just as well.

Seven steps from need to signed build. Vrisic analysis; timings are typical, not guaranteed

## What does SI software cost?

The same as AI software of the same scope. Vrisic’s typical ranges, with a fixed price after a free scoping call:

| Scope | USD | Timeline |
| --- | --- | --- |
| Single-task agent or workflow | $8,000 to $25,000 | 3 to 6 weeks |
| Integrated agent connected to your systems | $25,000 to $60,000 | 6 to 10 weeks |
| Multi-agent system | $60,000 to $150,000 or more | 10 to 16 weeks |

Running costs sit on top: hosting, model usage and monitoring, usually a small fraction of the build price per year but dependent on volume. See our [AI development cost guide](https://vrisic.com/blog/ai-development-cost/) for how budgets are built and [build vs buy](https://vrisic.com/blog/build-vs-buy-ai-software/) for when a ready-made product is the cheaper route. If someone quotes much more for the same scope because it is called SI, that is a premium for a word.

## How does this play out for different buyers?

These are illustrative scenarios, not client cases, showing how the same checks change by situation.

### A subcontractor bidding on a federal task order

The solicitation uses “SI” throughout and asks for a description of the proposed solution and how risk is managed. The bidder should mirror the solicitation’s terms in headings, add a one-line definition (“SI, as defined in the solicitation”) and put its real evidence up front: what the software does, test results on representative data, where data is hosted and how it is monitored. If the solicitation says nothing about a definition, the bidder can cite the order’s interim definition tied to 15 U.S.C. § 9401(3). It should not claim endorsement or “compliance” with the order, which sets no vendor requirements.

### A mid-sized private company buying an invoice or support agent

There is no reason to care about the label. The buyer’s checklist is scope, evidence on its own invoices or tickets, ownership, price and running cost. If a vendor’s proposal says SI, the buyer asks for a definition and the same ten answers. The contract should still carry the definition clause, because it costs nothing and prevents later arguments if marketing language shifts.

### A state or local agency

The order is addressed to federal departments and agencies, so a state agency follows its own procurement rules and style guide. Many will keep “AI” because state statutes and guidance use it. The practical move is the same: copy the wording of the solicitation, and write the scope in terms of what the software does.

## When should you not hire a development company at all?

Hiring a custom builder is not always the right answer. Skip it, or delay it, in these cases:

- **A standard product already does the job.** Common needs such as basic chat support, meeting notes or standard accounts payable often have mature software at a monthly price. Our [build vs buy guide](https://vrisic.com/blog/build-vs-buy-ai-software/) sets out the comparison.
- **You cannot describe the task.** If nobody can say what a good result looks like or provide 100 real examples, run an internal exercise first. Our AI readiness checklist helps.
- **The data is not available or not allowed.** If legal or security cannot approve the data flow, no vendor can fix that.
- **The need is a policy or strategy question.** A model cannot decide your risk appetite.
- **The goal is a label.** If the main driver is to say SI on a website, change the website, not the software.

## What is a low-risk process for selecting a vendor?

1. **Write the need in one page.** Outcome, volumes, systems, data, who approves, how success is measured.
2. **Shortlist three vendors** from different categories, for example a boutique, a mid-sized firm and a platform.
3. **Send the ten questions** and ask for written answers. Discard vague ones.
4. **Request a demo on a sample of your data.** Mask personal data first.
5. **Score with the sheet above** and discuss the gaps.
6. **Run a paid pilot of two weeks or less** with the top one or two, and credit the fee against the build.
7. **Sign a fixed-price build** with milestones, a warranty and a handover.

If you sell to the federal government, add one step: read the solicitation closely. If it defines SI, use that definition in your response. If it does not, state yours in one sentence and keep product descriptions factual.

## Where should you start your shortlist?

Our [top 10 AI and SI development companies in the USA](https://vrisic.com/blog/top-10-ai-and-si-development-companies-usa/) guide groups firms by buyer type. For broader lists see [best AI software development companies](https://vrisic.com/blog/best-ai-software-development-companies/) and [top AI agent development companies](https://vrisic.com/blog/top-ai-agent-development-companies/) , and for a framework see how to choose an AI development company. Our pages comparing Vrisic with other firms, such as Vrisic vs Toptal, are written openly as a competitor, so check their claims against the sources.

## How Vrisic can help

We build AI software, which US federal materials now call SI, for companies in the US, UK and UAE. We are a small remote team in India working US, UK and UAE hours. Price is fixed after a free scoping call, the system runs in your cloud account with your own model keys, you own the code, prompts and test data, and there are no per-seat fees. Read [how we work](https://vrisic.com/how-we-work/) and our [custom AI software development](https://vrisic.com/services/custom-ai-software-development/) service. We will tell you plainly if we are not the right fit.

To test before committing, the [AI Pilot](https://vrisic.com/ai-pilot/) costs $1,500 (£1,200, AED 5,500), takes 10 business days and is credited in full against a build started within 60 days. Message us on WhatsApp at +91 63777 67206 or [book a free scoping call](https://vrisic.com/contact/) .

## Sources

1. [Inaugurating the Era of Super Intelligence, Executive Order 14434](https://www.whitehouse.gov/presidential-actions/2026/09/inaugurating-the-era-of-super-intelligence/) (The White House, 29 September 2026)
2. [Fact Sheet: President Donald J. Trump Inaugurates the Era of Super Intelligence](https://www.whitehouse.gov/fact-sheets/2026/09/fact-sheet-president-donald-j-trump-inaugurates-the-era-of-super-intelligence/) (The White House, 29 September 2026)
3. [Inaugurating the Era of Super Intelligence, Federal Register document 2026-20321](https://public-inspection.federalregister.gov/2026-20321.pdf) (Office of the Federal Register, published 2 October 2026)
4. [15 U.S.C. chapter 119, National Artificial Intelligence Initiative](https://www.govinfo.gov/content/pkg/USCODE-2020-title15/html/USCODE-2020-title15-chap119.htm) (US Government Publishing Office)
5. [OMB issues memoranda on use and acquisition of AI by federal agencies, part 2](https://www.pilieromazza.com/omb-issues-memoranda-on-use-and-acquisition-of-ai-by-federal-agencies-part-2-what-it-means-for-government-contractors/) (PilieroMazza)
6. [Artificial Intelligence Risk Management Framework (AI RMF 1.0)](https://www.nist.gov/publications/artificial-intelligence-risk-management-framework-ai-rmf-10) (NIST)
7. [Superintelligence](https://en.wikipedia.org/wiki/Superintelligence) (Wikipedia)

Order details were checked on whitehouse.gov and the Federal Register in October 2026. Scoring weights, example vendors and cost ranges are Vrisic assumptions. Not legal advice.

## Frequently asked questions

### What is an SI development company?

It is a company that builds AI software for buyers who use the federal term Super Intelligence. Under Executive Order 14434, SI means the technologies covered by the statutory definition of artificial intelligence in 15 U.S.C. § 9401(3), so the software is the same: agents, voice assistants, document processing and search over your own files. The label does not change what is being built.

### Can a company build true super intelligence?

No company has shown a system that greatly exceeds human performance across virtually all domains, which is the technical meaning of superintelligence. The federal use of SI does not require it, because the order ties SI to the existing AI definition. Treat any vendor claiming to sell beyond-human intelligence as a red flag and ask for measured results on your own data.

### Does the SI executive order apply to private companies?

Not directly. Executive Order 14434 directs executive departments and agencies to use SI and SI wording in their non-statutory documents, and it creates no enforceable rights for any party. It does not mention private companies. You may meet the wording if you sell to a federal agency, so follow each solicitation, and otherwise you can keep saying AI.

### How much does SI software cost compared with AI software?

The same. Price depends on scope, not the label. Vrisic’s typical ranges are $8,000 to $25,000 for a single-task agent, $25,000 to $60,000 for an integrated agent and $60,000 to $150,000 or more for a multi-agent system, plus hosting and model costs. If a vendor quotes a premium just for calling it SI, ask what extra work is included.

### How should I define SI in a contract?

Tie it to something stable, for example: “AI or SI means the technologies covered by 15 U.S.C. § 9401(3), unless the buyer’s solicitation or a later law defines it otherwise.” The order also asks for a proposed legislative definition within 60 days, so allow for change by referring to the buyer’s documents. Have counsel review the clause.

### What should I ask a vendor before hiring them?

Ask what the software does on day one, how accuracy is tested on your data, the fixed price, who owns the code, prompts and keys, where data is stored and whether it is used for training, how mistakes are handled, monthly running cost and how they define SI. Then request a demo on masked data and, where evidence is thin, a short paid pilot credited against the build.

### Should I use SI or AI on my company website?

Cover both terms. Customers still search for AI, while federal buyers and contractors may use SI. Lead with outcomes such as invoices processed or calls answered, add a short glossary line, and follow the buyer’s wording in proposals. Do not imply government approval or capabilities beyond human ability, because the order does not endorse any vendor or product.

---
Vrisic · https://vrisic.com/ · Contact: info@vrisic.com · Book a call: https://vrisic.com/contact/