DFX Intelligence · reference
The checkable version
Everywhere else on this site is trying to convince you. This section is trying to be accurate. It is the reference material a buyer, an analyst, a journalist or a machine needs in order to describe the General correctly, including the parts where something else is the better choice.
Start here
Capabilities
One page per thing the General actually does. Each one states the problem, the mechanism, what you can check for yourself, the limits, and when to choose something else.
AI that researches a prospect and writes the first approach
The General reads public sources about a specific company and person, decides whether they actually fit the brief, writes an approach grounded in what it found, and then owns the follow-up sequence.
AI that uses a real browser on your own accounts
The General drives a real Chrome session rather than an API, so it can work inside portals, directories and tools that have no API and would otherwise need a person.
AI that attends the meeting and then does what was agreed
The General joins the call, keeps the recording, transcript and write-up, and then turns what was committed to on the call into actual queued work rather than a list of bullet points.
AI that works your inbox and never drops a follow-up
The General reads the mailbox you already have, triages what actually needs you, drafts what does not, and keeps chasing the threads that went quiet.
AI that answers the phone and books the job
The General answers a real phone line, handles the conversation, and completes the thing the caller rang about, then writes the outcome into your systems.
AI that keeps the CRM current without data entry
The General writes what it did into the CRM you already have, so the record reflects the work rather than somebody's memory of it.
Task documentation
How to do a specific thing, rather than what a capability is. One question per page, answered in the first paragraph, with the prerequisites and the limits stated on the page rather than discovered later.
How do I get an AI to join my meeting and actually send the follow-up?
Invite the General to the call the way you would invite a person. It joins as a participant, keeps the recording, transcript and written recap, and turns what was committed to on the call into queued work. The follow-up email is drafted from what was actually said and waits for your approval before it sends.
How do I stop doing CRM data entry after every call?
Connect the CRM you already use. The General writes what it did into it as the work happens, with the evidence attached, so the record reflects the work rather than somebody's memory of it at the end of the day.
How do I make sure an automated follow-up stops the moment someone replies?
It stops on its own. A reply moves the record and cancels the remaining sequence, because the General is reading the mailbox the sequence sends from rather than tracking opens from outside it.
How do I get an AI to research a company properly before it emails them?
Describe the buyer the way you would to a new salesperson rather than as filter criteria. The General reads public sources about the specific company and person, decides whether they actually fit that brief, and writes an approach that turns on what it found.
How do I get an AI to work LinkedIn and email together without risking my account?
The General works LinkedIn through a real Chrome session you are already signed into, rather than through an unofficial API. Your credentials are never sent to DFX, because it acts inside the session on your own machine, and email runs from your own connected mailbox alongside it.
Comparisons
Every comparison here contains at least one criterion the alternative wins, and a real recommendation to choose it. That is enforced by a test, not by good intentions.
The General vs hiring a sales development rep
An SDR is better when the job needs judgement in a live conversation and you can afford the ramp; the General is better when the constraint is that nobody is doing the follow-up at all.
AI agent vs AI copilot: what the difference actually is
A copilot makes the person at the keyboard faster; an agent removes the requirement that somebody is at the keyboard. Neither is better in general and they fail in opposite ways.
The General vs a workflow automation tool
Use a workflow tool when the steps are known and stable; use an agent when deciding the steps is the work.
AI notetaker vs an AI that acts on the meeting
A notetaker is a better archive and costs less; an acting agent is worth it only if the failure you are fixing is that nothing happens after the call.
The General vs a browser-only agent
A browser agent completes a task in a tab; an operator owns an outcome across several surfaces and keeps going tomorrow.
Original research
Research DFX produced because we ran the measurement, with the method, the sample and the limitations published alongside the finding. Where a dataset exists, it is downloadable.
2026-08-19
Which sources AI answer engines cite when buyers ask about business software
For commercial software buying questions, AI citations are extraordinarily fragmented, not concentrated. Across 2,391 citations we recorded 1,330 distinct domains, 72% of which were cited exactly once, and the top 25 domains together accounted for 14.4% of all citations. Reddit, the domain most often described as the single biggest source in AI answers, was 0.63%. Wikipedia was 0.33%. Capterra was 0.04%.
2026-08-19
How often AI assistants actually search the web before answering a buying question
20.8% of answers to commercial buying questions involved no web retrieval at all, and it depends enormously on how the question is phrased: 97.4% of category questions ("what are the best AI sales agents") triggered a search, against 50.0% of comparison questions ("X versus Y"). The answers that skip retrieval are drawn entirely from what the model already believes about the market, and no amount of work on your own website reaches them.
The Observatory
Original visual intelligence on business, capital, artificial intelligence, property and labour, published at /observatory. Every figure traces to a named primary source, every dataset ships the verification checks it passed, and everything is reusable under CC BY 4.0 with attribution.
2026-08-24
America files twice as many new businesses as it did in 2019. Most of them will never hire anyone.
Business applications are up 77% in seven years. The states where they grew fastest are the states where the smallest share look like real employers.
2026-08-24
Artificial intelligence is now in the majority of American annual reports
Seven years ago one annual report in eighteen mentioned AI. The number of annual reports barely changed. What companies feel obliged to address did.
2026-08-24
Wyoming has the fastest business formation in America and the slowest economic growth
Filings and output move together in almost every state. Removing Wyoming from the data makes the relationship stronger, which is the clearest evidence yet for what its filings actually are.
CSV · 12 rows · tier A
Share of SEC annual reports mentioning artificial intelligence
Source: U.S. Securities and Exchange Commission, EDGAR full-text search. Free to cite with attribution.
CSV · 51 rows · tier A
Business formation by state, trailing twelve months
Source: U.S. Census Bureau, Business Formation Statistics. Free to cite with attribution.
CSV · 25 rows · tier A
DFX Business Weather: temperature and momentum for 25 US metros
Source: U.S. Bureau of Labor Statistics (employment, unemployment, labour force) and U.S. Census Bureau (building permits). Free to cite with attribution.
CSV · 5 rows · tier A
Signals: things worth knowing today
Source: DFX Intelligence Observatory, derived from the datasets each signal cites. Free to cite with attribution.
CSV · 51 rows · tier A
Business formation against real economic growth, by state
Source: U.S. Bureau of Economic Analysis, Regional Economic Accounts (SAGDP9), with U.S. Census Bureau Business Formation Statistics. Free to cite with attribution.
Business Weather
A temperature and momentum reading for 25 US metro economies, rebuilt from public BLS and Census series and dated to 2026-07. Each market is scored against its own history. It describes current conditions and does not forecast, which was tested rather than assumed.
Machine-readable
The same facts, as data. These exist because a person building something against our public record should not have to scrape a page to get it, and because a record that is served in two formats from one source cannot disagree with itself.
JSON
/ai/entity.json
The organisation, the product, the definition, the plans and the limits, dated.
JSON
/ai/capabilities.json
Every capability with its problem, mechanism, evidence and limits.
MCP SERVER
Real Estate MCP server
Machine-callable US property, parcel, ownership, recorded sale and CRE debt-maturity intelligence. Seven tools, no authentication, free. Coverage numbers and gaps stated.
The General by DFX Intelligence. DFX Intelligence is the operating brand for the General platform. It is not affiliated with DFX Holdings Group, with Deal Flow Xchange, or with any audio-processing product using the DFX name. This section was last verified on 2026-08-19.